THE ECONOMICS OF MOBILE INTERNATIONAL ROAMING · PDF fileTHE ECONOMICS OF MOBILE INTERNATIONAL...

THE ECONOMICS OF MOBILE INTERNATIONAL

ROAMING

by

Saeed Alkatheeri

Thesis submitted for the degree of Doctorate of Philosophy

University of East Anglia

School of Economics 31/12/2013

This copy of the thesis has been supplied on condition that anyone who consults it is understood to recognise that its copyright rests with the author and that use of any information derived there from must be in accordance with current UK Copyright Law. In addition, any quotation or extract must include full attribution.

2

SAEED ALKATHEERI 2013 THE ECONOMICS OF MOBILE INTERNATIONAL ROAMING

Abstract

International roaming is a hot topic in the telecommunications industry. Many countries have witnessed a downward trend in mobile domestic prices. On the contrary, international roaming prices remained reluctant to follow the domestic trend. In Europe, the service has been regulated with price cap since 2007, and regulation is maintained for years to come. The existing literature on the economics of international roaming has focused on theoretical modelling, which assumes a uniform retail price (i.e. common across visited networks). The main finding is that wholesale and retail prices rise with the number of visited networks. Additionally, vertical merger is found unprofitable; and home network steering does not cause downward pressure on wholesale prices. We found that the assumption of uniform retail pricing leads to results that are inconsistent with wholesale competition because visited networks appear in the demand as complements rather than substitutes. We present theoretical models that match the existing literature’s findings, and compare results to the case whereby the retail price is discriminatory (i.e. differs by visited networks). With discriminatory retail, substitutability of networks reduces prices, and the incentive for vertical merger exists. In a steering game, steering is found able to reduce wholesale prices; and networks alliances are formed in equilibrium. The empirical literature on international roaming is limited to few industry studies. We use an aggregated dataset on prices and quantities for networks visited by roamers from one major mobile provider whose subscribers travel a lot across the world, Etisalat. The study period witnessed a retail price shift from discriminatory to uniform. The main findings are: (1) competition, as measured by the number of visited networks, reduces wholesale price; (2) traffic steering is effective, especially towards preferred networks (alliance and cross-owned); (3) only alliance networks offer wholesale discounts; and (4) demand is more elastic than crude industry studies.

3

ACKNOWLEDGMENTS

I am grateful to the Telecommunications Regulatory Authority in the United Arab Emirates,

the Board members, and the Director General for privileging me with the PhD scholarship. I

gratefully acknowledge the fruitful suggestions of the Regulatory Affairs team and their help

in my data collection. I also thank Etisalat for providing real data for analysis in this thesis.

I am greatly indebted to Professor Bruce Lyons for his supervision and guidance. I

appreciate all his efforts and time he spent with me throughout the work of this thesis.

I am deeply thankful to Dr. Subhasish Modak Chowdhury and Dr. Farasat Bokhari for their

support and advice regarding the models in the thesis.

My thanks are extended to all staff and students of the University of East Anglia whom I

interacted with and learned from in my MPhil upgrade and during all the four years of my

research degree.

I also thank Dr. George Symeonidis and Dr. Sang-Hyun Kim for being the examiners of my

thesis and for providing valued suggestions and corrections for the thesis.

I recognize the support of the NRA in the Sultanate of Oman in providing me with retail price

information regarding the Omani market. I also recognize the support of Cullen International

for providing me with access to Cullen International regulatory intelligence services.

Last but not least, I thank all my family members, with deepest gratitude towards my wife

and our children for their constant support; and towards my parents for being the sources of

my inspiration: my mother, and my late father whose memory will be with me always.

4

TABLE OF CONTENTS

Chapter (1). Introduction ..................................................................................................................... 10

1.1 The Service Of Mobile International Roaming ..................................................................................... 11

1.2 Motivation ................................................................................................................................................ 15

1.3 Outline Of Thesis ..................................................................................................................................... 16

Chapter (2). Literature Review ............................................................................................................ 18

2.1 International Roaming In EU ................................................................................................................. 18 2.1.1 Sector Inquiry ................................................................................................................................. 18 2.1.2 Aftermath Of Sector Inquiry ........................................................................................................... 20 2.1.3 EC Roaming Regulation ................................................................................................................ 23

2.2 Vertically Related Industries .................................................................................................................. 23 2.2.1 Access Theory ............................................................................................................................... 25 2.2.2 Perfect Complementarity ............................................................................................................... 28

2.3 Literature On International Roaming ................................................................................................... 31 2.3.1 Published Papers ........................................................................................................................... 31 2.3.2 Working Papers ............................................................................................................................. 34 2.3.3 Empirical Literature ........................................................................................................................ 36

2.4 Discussion ................................................................................................................................................. 40

Chapter (3). The Base Model ............................................................................................................... 44

3.1 Model Background .................................................................................................................................. 45 3.1.1 Demand Side ................................................................................................................................. 45 3.1.2 Supply Side .................................................................................................................................... 46 3.1.3 Pricing ............................................................................................................................................ 48

3.2 The Game ................................................................................................................................................. 48 3.2.1 Players ........................................................................................................................................... 48 3.2.2 Demand Specification .................................................................................................................... 49 3.2.3 Payoffs ........................................................................................................................................... 52 3.2.4 Timing Of The Game ..................................................................................................................... 53 3.2.5 Wholesale Pricing Strategies ......................................................................................................... 53 3.2.6 Retail Pricing Policies .................................................................................................................... 54 3.2.7 Stage II .......................................................................................................................................... 55 3.2.8 Stage I ........................................................................................................................................... 57

3.3 Results ....................................................................................................................................................... 63 3.3.1 Response Functions ...................................................................................................................... 63 3.3.2 Payoffs Comparison ....................................................................................................................... 64 3.3.3 Welfare Implications ....................................................................................................................... 67

3.4 Discussion ................................................................................................................................................. 68

Chapter (4). Preferential Wholesale Agreements ............................................................................... 71

4.1 Cross-Ownership Model .......................................................................................................................... 71 4.1.1 Model Background ......................................................................................................................... 72 4.1.2 Discriminatory Retail Policy ........................................................................................................... 73 4.1.3 Uniform Retail Policy ...................................................................................................................... 78 4.1.4 Concluding Remarks ...................................................................................................................... 82

4.2 Uniform Price With Asymmetric Market Shares ................................................................................. 84

4.3 Roaming Alliance Model ......................................................................................................................... 86 4.3.1 Model Background ......................................................................................................................... 87 4.3.2 Steering Investment Game ............................................................................................................ 89 4.3.3 Roaming Alliance Formation Game ............................................................................................... 90 4.3.4 Concluding Remarks ...................................................................................................................... 93

5

4.4 Discussion ................................................................................................................................................. 94

Chapter (5). Dataset Description ......................................................................................................... 97

5.1 Relevance Of Dataset ............................................................................................................................... 97

5.2 Brief On Etisalat Retail Prices ................................................................................................................ 98

5.3 Data Collection ....................................................................................................................................... 101

5.4 Information On Foreign Networks ....................................................................................................... 102

5.5 The Dataset ............................................................................................................................................. 106

5.6 Discussion ............................................................................................................................................... 114

Chapter (6). Demands For Visited Networks .................................................................................... 115

6.0 Simple Logit Demand ............................................................................................................................ 116

6.1 The Dataset ............................................................................................................................................. 117 6.1.1 Visited Networks Characteristics.................................................................................................. 119 6.1.2 Variables ...................................................................................................................................... 120 6.1.3 Sampling Problem ........................................................................................................................ 122

6.2 Empirical Methodology ......................................................................................................................... 122 6.2.1 Market Definition .......................................................................................................................... 123 6.2.2 Market Size .................................................................................................................................. 124 6.2.3 Estimation Model ......................................................................................................................... 125 6.2.4 Endogeneity Problem ................................................................................................................... 126

6.3 Results And Discussion .......................................................................................................................... 131

6.4 Concluding Remarks ............................................................................................................................. 136

Chapter (7). Visited Networks Wholesale Pricing............................................................................. 138

7.1 Empirical Methodology ......................................................................................................................... 139



Chapter (8). Effectiveness Of Steering .............................................................................................. 149

8.1 Empirical Methodology ......................................................................................................................... 150



Chapter (9). Conclusion ..................................................................................................................... 159

Bibliography ....................................................................................................................................... 165

Appendix (A). Choice On Retail Policy By A Monopolist ................................................................ 169

Appendix (B). Derivations Of Base Model Equilibria ...................................................................... 171

Appendix (C). Networks Choice On Retail Policy ............................................................................ 181

Appendix (D). Instrumental Variables .............................................................................................. 185

Appendix (E). Additional Estimations For Demand And Steering Models ..................................... 191

Appendix (F). Mean Elasticities In EEA Countries ......................................................................... 200

Appendix (G). List Of Visited Networks ............................................................................................ 201

6

LIST OF TABLES

Table (2.1) NRAs’ responses to the EC market analysis notification exercise for Market (17). ............................................... 22 Table (2.2) Illustration of perfect complementarity at different values of . .......................................................................... 30 Table (3.1) Standardised wholesale and retail prices under discriminatory and uniform retail policies. ................................ 64 Table (4.1) Payoffs matrix for joining a cross-ownership pair under the discriminatory retail policy. .................................... 77 Table (4.2) Payoffs matrix for joining a cross-ownership pair under the uniform retail policy. ............................................... 81 Table (4.3) Standardised prices under discriminatory and uniform retail policies in the cross-ownership game. .................. 82 Table (4.4) Payoffs matrix for joining a roaming alliance. ........................................................................................................ 93 Table (4.5) Standardised prices when all networks apply the monopoly retail price in the steering investment game. ........ 94 Table (5.1) Etisalat international roaming average retail prices in AED for postpaid roamers in 2009. .................................. 99 Table (5.2) Etisalat international roaming retail prices in AED for postpaid roamers in 2010. .............................................. 100 Table (5.3) Etisalat roaming alliance and cross-owned visited networks by country. ........................................................... 104 Table (5.4) Number of visited networks with CAMEL technologies by country as of 2010. .................................................. 105 Table (5.5) Visited networks with Multinetwork by country. ................................................................................................ 106 Table (5.6) Dataset summary statistics. ................................................................................................................................. 109 Table (5.7) Correlations between roaming services. ............................................................................................................. 112 Table (6.1) Notations and variables related to estimations. .................................................................................................. 121 Table (6.2) Calculations related to market size. ..................................................................................................................... 125 Table (6.3) Simple logit demand models. .............................................................................................................................. 133 Table (6.4) Marginal effects on visited networks’ characteristics post Etisalat uniform retail policy. ................................... 135 Table (7.1) Notations and variables related to estimations. .................................................................................................. 140 Table (7.2) Wholesale price models with wholesale price in level-form. .............................................................................. 143 Table (7.3) Wholesale price models with wholesale price in log-form for all observations. ................................................. 144 Table (7.4) Marginal effects on visited networks’ characteristics post Etisalat uniform retail policy. ................................... 145 Table (8.1) Notations and variables related to estimations. .................................................................................................. 151 Table (8.2) Market share models for all observations. .......................................................................................................... 154 Table (8.3) Marginal effects on networks’ characteristics and relative prices post Etisalat uniform retail policy. ................ 155 Table (8.4) Hypothetical example of wholesale undercutting in a market with three visited networks. .............................. 155 Table (D.1) Instrumental variables for simple logit demand models with Oman-Mobile. ..................................................... 187 Table (D.2) Tests for the instrumental variables used in Table (D.1). .................................................................................... 188 Table (D.3) Instrumental variables for simple logit demand models with average Oman-Mobile. ....................................... 189 Table (D.4) Tests for the instrumental variables used in Table (D.3). .................................................................................... 190 Table (E.0) Possible scenarios for the hypotheses test results and our intuitions. ................................................................ 193 Table (E.1) Notations and variables related to estimations. .................................................................................................. 194 Table (E.2) Simple logit demand models for observations in Highly Visited Markets. ........................................................... 195 Table (E.3) Simple logit demand models for observations in Low Visited Markets. .............................................................. 196 Table (E.4) Simple logit demand models for all observations. ............................................................................................... 197 Table (E.5) Marginal effects on visited networks’ prices post Etisalat uniform retail policy. ................................................ 198 Table (F) Estimated mean own price elasticity of demand for EEA visited networks by market. .......................................... 200 Table (G) List of all foreign networks visited by Etisalat roamers during the years 2008-2011. ............................................ 201

7

LIST OF FIGURES Figure (2.1) Prices of perfect complementary firms at different values of n. ...................................................... 30 Figure (3.1) International roaming in two countries each with two networks. .................................................... 49 Figure (3.2) The demand facing firm 1. ................................................................................................................ 51 Figure (3.3) Equilibrium profit to network in each IOT agreement. ................................................................... 65 Figure (5.1) Map for highly visited markets by Etisalat roamers during the years 2008-2011. .......................... 110 Figure (5.2) Wholesale price for a minute of outgoing call by GCC visited networks by year. ........................... 111 Figure (6.1) Own price elasticity by year. ........................................................................................................... 134 Figure (C) Networks choice on retail policy in extensive form. .......................................................................... 184

8

LIST OF ACRONYMS

AED UAE currency Dirham; pegged to US dollar (1 AED=$0.27)

CAMEL Customized Applications for Mobile Enhanced Logic

CPI Consumer Price Index

EC European Commission

EEA European Economic Area

EU European Union

GCC Gulf Cooperation Council (member states: Bahrain, Kuwait, Oman, Qatar, Saudi Arabia and the UAE)

GDP Gross Domestic Product

IOT Inter-Operator Tariff (wholesale price used in mobile international roaming)

MB Megabytes of data

NBS The National Bureau of Statistics in the UAE

NRA National Regulatory Authority

OECD The Organisation for Economic Co-operation and Development

OLS Ordinary Least Squares estimator

POSTPAID Mobile Monthly Plan Subscription

PREPAID Mobile Pay-As-You-Go Subscription

SMS Short Message Service (text message)

TAP Transferred Account Procedure code

UAE United Arab Emirates

1SLS First Stage Least Squares estimator

2SLS Second Stage Least Squares estimator

9

Dedicated To

My Wife

And

Our Children

10

Chapter (1). Introduction

When two mobile networks licensed in two different countries sign a bilateral Inter-Operator

Tariff (IOT) agreement, their subscribers are allowed to roam across the networks with their

phone handsets to make or receive phone calls and text messages (SMS), and to access

mobile wireless internet.

Many countries have witnessed a declining trend in most mobile domestic prices due to

different reasons such as competitive pressures, price regulation and technological

improvements that increased networks efficiency. On the contrary, international roaming

prices had increased and remained reluctant to follow the domestic trend. Inspired by this

phenomenon, we are interested in understanding how networks would set their wholesale

prices under their bilateral IOT agreements, which also affect their retail prices.

The existing literature assumes the roamer only cares about the average retail price, and

hence the retail price used in the demand is uniform. The home network marks up the

average wholesale price. We introduce in the modelling the assumption of retail prices that

differ by visited networks (i.e. discriminatory). The assumptions made on retail pricing

(uniform versus discriminatory) give opposite predictions with regards to competition and the

incentives for networks to vertically merge.

The empirical literature on wholesale competition is limited to few industry studies. We make

use of an aggregated dataset on prices and quantities for networks visited by roamers from

one major mobile provider whose subscribers travel a lot across the world, Etisalat. The

study period records a shift in Etisalat retail pricing policy (from discriminatory to uniform),

which provides an empirical experiment to measure wholesale competition.

This chapter is organized as follows. Section (1.1) explains international roaming in the

context of mobile services. Section (1.2) provides thesis motivations, and Section (1.3)

outlines the next chapters.

11

1.1 The Service Of Mobile International Roaming

If a mobile network is closed, its subscribers can only communicate within the network’s

subscribers. If the network interconnects with other networks (i.e. signs

interconnection/wholesale agreements), it enables its subscribers to communicate with

subscribers to those networks. Interconnection agreements therefore make networks

compatible.

Interconnection agreements can be classified into two types: one-way access, whereby only

one network needs to access the facility of another network; and two-way access whereby

the two networks need access to one another’s facilities. The agreeing networks can be

competing within the same geographic market, or operating in different geographic markets.

Mobile-to-mobile (MTM) call termination access for off-net calls and national roaming tariffs

are examples of wholesale prices of interconnection agreements within a geographic market.

In parallel, international settlement rate agreements for terminating international calls and

international roaming tariffs (IOT) are examples of wholesale prices of cross-border

interconnection agreements. If interconnection charges are per unit of use, they act as

“perceived marginal cost” to the network that incurs them.

This thesis concentrates on international roaming services governed by IOT agreements that

allow a home network subscriber to use his same phone number while using (roaming on) a

foreign network1. IOT agreements are typically a bilateral (two-way access) agreement.

Despite international roaming similarities with other telecommunications services that involve

interconnection agreements, international roaming has some features that make it unique,

as explained below.

The consumer of international roaming has to be in a different geographic market. He

effectively borrows the facilities of a foreign network to originate or terminate traffics (e.g.

making an outgoing call while roaming). The IOT agreement is between non-rivals, which is

similar to the international settlement rate agreement for international calling.

1 For a technical definition, see GSMA (2013).

12

With national roaming, the consumer usually does not pay a premium for any mobile service

when roaming on a different network than his home network within the same market. In other

words, retail prices are the same whether the consumer uses his home network or any other

domestic network. In addition, in areas where the home network provides coverage, the

consumer cannot select manually (on his handset) between networks other than his home

network. Given no price differentials, consumers typically leave handsets on automatic mode

to be picked up by the home network or by its national roaming partners in areas the home

network has a lack of coverage.

On the contrary, with international roaming, the consumer is usually aware that the traffic he

uses costs a premium2 when roaming in a foreign land, and he will be charged for incoming

calls3. Paying for incoming calls (rather than free) is supposed to eliminate consumer’s

substitutability in call making-receiving4. In addition, the consumer can select manually (on

his handset) between visited networks that are IOT partners of his home network. In other

words, the consumer can choose a wholesale supplier to his home network. We will later

assume the consumer leaves his handset on automatic mode if there are no price

differentials.

The home network collects retail roaming revenues from its travelling subscribers, and pays

IOTs (i.e. wholesale prices) to visited networks5 for the use of their facilities. The retail price

is typically the IOT plus a markup6.

Retail prices for international roaming are generally classified into four categories which also

represent the wholesale services in a typical IOT agreement: (i) making an outgoing call

(destined to local phone number within the visited country or to the home country or a third

country); (ii) receiving a call from any destination; (iii) sending a SMS to any destination7;

and (iv) using megabytes (MB) of data roaming.

In order to enjoy international roaming, the consumer has to be a subscriber to a home

network, and this home network has an IOT agreement with visited networks. In order for his

demand to be satisfied, the roamer has to be in an area where a visited network provides

coverage. In populated areas, where roamers would typically visit, visited networks would

2 Of course, the consumer can quit roaming and use outside options instead, such as using a visitor SIM-card or public payphones. 3 This is true even in countries that follow the Calling Party Pay (CPP) regime. 4 Hakim and Lu (1993) modelled international calling (under international settlement rate agreement) assuming making the call and receiving it as near-perfect substitutes. 5 Mobile operators use TAP as the billing standard, by which a visited network sends billing records of roaming subscribers to their respective home network (GSMA, 2013). 6 Some countries impose value-added tax (VAT) on wholesale or/and retail prices. 7 Incoming SMS is free.

13

tend to have coverage overlaps that should enable roamers to substitute between visited

networks.

Roamers have two options at their handsets to select a visited network: automatic or manual

network selection. The automatic mode is usually the default setting in mobile handsets,

which would be typically picked up by the network with the strongest coverage signal.

Leaving a handset on the default selection could be due to a number of reasons: (1) the

roamer’s unawareness of manual selection; (2) unawareness of retail price differential

between visited networks (discriminatory); or (3) the roamer has no preference over the

visited networks. The automatic mode of handsets is a necessity for home network’s traffic

direction technologies (i.e. steering) to take effect.

The rational roamer is supposed to manually select between visited networks if this selection

raises his utility, such as reducing his expenditure. The roamer’s manual selection is a

demand-side substitution, which obstructs steering made by his home network (supply-side

substitution). One way to make the roamer indifferent in the selection between visited

networks is by charging him a uniform retail price.

The roamer with prepaid subscription can enjoy international roaming if both his home

network and the visited network installed Customised Applications for Mobile Enhanced

Logic (CAMEL) technology to allow the home network to instantly deduct the credits

according to usage. At the wholesale level, a visited network charges an IOT price to a home

network with no price discrimination between the home network’s (prepaid/postpaid)

subscribers or their place of visit within the visited market8 . At the retail level, prepaid

subscribers tend to be charged more for roaming compared to postpaid.

Mobile networks may involve in preferential IOT agreements in the form of cross-ownership

or roaming alliance membership, which result in preferential wholesale prices, and lower

retail prices when roaming on a member’s network. Vodafone and ten of its affiliates in

Europe notified the European Commission (EC) in 2001 to set preferential wholesale prices

and to offer a pan-Europe retail price (Eurocall) for roaming on Vodafone networks (O.J.

2001).

8 Networks located near borders may unintentionally cause accidental roaming. This happens if two networks are IOT partners and the subscriber to any of them is at his home country but near the border. If the foreign network has a stronger signal with coverage stretching over the subscriber whose handset is on automatic mode, the subscriber unintentionally consumes international roaming units and the foreign network charges IOT bills to the home network. Accidental roaming is usually solved through consumer complaints.

14

At the retail level, Vodafone’ Eurocall is lower for roaming on its’ cross-owned networks, but

differs by visited networks. The practise of the home network setting retail prices that are

discriminatory (by visited networks) used to be the common practise in this industry. Many

mobile networks shifted from this discriminatory policy to a uniform policy9 (i.e. identical retail

price across visited networks in the same market), despite being charged different wholesale

prices by the visited networks, as noted by Ambjørnsen, et al. (2011).

The practise of uniform retail policy and the formation of roaming alliances are also observed

outside Europe. In the UAE, the mobile operators Etisalat and Du used to charge

discriminatory retail prices. Du started offering uniform retail prices in 2008 (EITC, 2008);

while Etisalat in 2010 started uniform prices based on geographic zones (Gulfnews.com,

2010).

In an IOT agreement, each mobile operator serves two distinct demands: (1) own

subscribers at the retail level (or outbound) and (2) visitors at the wholesale level (or

inbound). Any cross-subsidisation between the two groups (outbound and inbound roamers)

can be a response to competition on one side, but there is no linkage between the two

groups. That is, there is no pricing structure for the two groups that can lead to an inter-

group network externality. Hence, international roaming is not a two-sided market (Shortall,

2010). We shall see in the next chapters that the wholesale and retail demands for a network

are independent.

Despite competition for subscribers within a market, a mobile network holds monopoly power

in providing access to its subscriber because any communication with him has to come

through it. This competitive bottleneck, as in Armstrong and Wright (2009), is reinforced after

crossing borders, where different interconnection agreements may intersect to serve the

travelling bottleneck.

For example, consider a voice call by A1’s subscriber to A2’s subscriber, where A1 and A2

are mobile networks in country A. Let us suppose that A2’s subscriber happens to be

roaming on foreign network B. In this case, A1’s subscriber pays local (off-net) retail price to

A1; A1 pays mobile-to-mobile (MTM) domestic access (wholesale) to A2 for terminating the

call; A2 charges its travelling subscriber a retail price for incoming call while roaming; and A2

pays B an international settlement rate (wholesale) for the international leg of the call plus an

9 In Appendix (A), we show discriminatory pricing is superior to uniform pricing for a monopolist selling two substitutable goods with different (actual) marginal costs, ignoring upstream strategic behaviours for simplicity. This conclusion is more appealing as the two goods become more asymmetric in marginal costs or/and more substitutable.

15

IOT (wholesale) price for the incoming call. In this example, three interconnection

agreements are involved: MTM access, international settlement rate and IOT.

Nonetheless, interconnection agreements are complementary. An IOT agreement does not

replace non-IOT interconnection agreements since they are typically signed independently

and billed to end users in different transactions (Shortall, 2010). Moreover, IOT agreements

are voluntary agreements, similar to international settlement rates 10 , and are widely

unregulated compared to other interconnection agreements.

1.2 Motivation

Despite the rise in the number of mobile networks overtime, technological improvement, and

decrease in domestic prices, prices of international mobile roaming services are exceptional

to the observed price trend in the mobile industry. In Europe, the EC regulated international

roaming by imposing price caps for roaming within the European Economic Area (EEA)

countries. This phenomenon inspired the theoretical modelling of Salsas and Koboldt (2004),

Lupi and Manenti (2009), Ambjørnsen, et al. (2011), and some others as shall be surveyed

in Chapter (2).

The gap in the theoretical modelling of wholesale competition lies in the assumption of a

uniform retail price. By assuming a uniform price, the demand reflects visited networks as

perfect complements rather than substitutes. This will be elaborated in the next chapters,

and will become clear as we compare price equilibria to the case of discriminatory retail

prices.

Contrary to the existing literature, we provide a game for home network steering investment

and roaming alliance formation, assuming uniform retail price that takes into account

substitutability of visited networks.

In addition, IOT agreements are rarely regulated as opposed to most wholesale agreements,

making it interesting to undertake empirical research on market outcomes. We are not aware

of any empirical paper that examines competition in international roaming.

10 Cave and Donnelly (1996) describe international settlement agreements as voluntary, which satisfies Edgeworth’s requirement: any agreement must make both parties at least as well off as no agreement (individual rationality) and any agreement must be Pareto optimal.

16

We are provided with access to data on Etisalat’s outbound roaming at the visited network

level, observed over four years. Etisalat retail prices were discriminatory in the first two

years, then uniform in the rest of years. Such data will help in testing the theoretical

predictions regarding the impact of the retail policy and preferential IOT partners on

wholesale competition and on market share to explore for steering. Furthermore, the data

will help in the estimation of the demands for visited networks.

The thesis focuses on the economics of mobile international roaming because of the

aforementioned gaps in both the theoretical and empirical literatures. We draw attentions to

the implications of home network retail policy (discriminatory versus uniform) on consumers’

choice between automatic and manual network selection, and on visited networks’ wholesale

prices and steering by the home network.

1.3 Outline Of Thesis

The next chapters are organized as follows. Chapter (2) reviews the literature. It starts with

the EU experience with international roaming as a competition case. The chapter

summarizes: the main findings of the EC’s sector inquiry, competition analyses by European

NRAs, and the EC’s intervention in the industry. Then the chapter reviews the literature on

vertically related industries, and reviews the existing papers on international roaming.

Finally, the chapter places our next chapters in the context of the existing literature.

Chapter (3) lays down the base model. The chapter explains the underlying assumptions

used in modelling the behaviours of networks at the retail and wholesale levels. Then

different wholesale pricing strategies are explored under discriminatory and uniform retail

pricing policies. Finally, results are discussed.

Chapter (4) extends the base model to address preferential IOT agreements. The chapter

examines the incentive for networks to engage in vertical mergers, and compares results

under discriminatory and uniform retail pricing policies. Finally, a steering game is presented

to understand the incentive for steering investment and the formation of roaming alliances.

Chapters (5) describes the dataset of Etisalat outbound roaming. The chapter explains the

relevance of the dataset with regards to understanding wholesale competition, and explains

the data gathering process. Finally, the dataset variables are explained and summarized.

17

Chapter (6), (7) and (8) contain the empirical estimations, using the dataset on Etisalat

outbound roaming. Chapter (6) estimates the demands by Etisalat roamers for visited

networks. Chapter (7) estimates the visited networks’ wholesale prices to Etisalat. The

chapter takes into consideration the visited networks characteristics which may reflect

preferential IOT agreements, and market structure that may reflect competition effects.

Chapter (8) estimates visited networks’ market shares to explore for the effectiveness of

steering by Etisalat.

The Appendices at the end contain: (A) a model for a monopolist choosing between

discriminatory and retail policies; (B) the derivations of the base model’s solutions; (C) a

game for networks choosing between discriminatory and uniform retail policies; (D) the

evaluations of instrumental variables used in demand estimations; (E) the estimated own

price elasticities of demands for EEA networks; and (F) the list of all foreign networks visited

by Etisalat roamers during the study period.

18

Chapter (2). Literature Review

This chapter reviews the literature related to Inter-Operator Tariff (IOT) agreements between

mobile network operators, which govern international roaming.

We start with a brief history on international roaming under the EC legal framework (Section

2.1). In Section (2.2), we provide a short survey on vertically related industries that

encompasses interconnection agreements. Specifically, we review three papers on

international settlement rates due to its similarity to IOT agreements; and a paper on fixed-

to-mobile network access as it is referred to in some papers that model IOT agreements. In

Section (2.3), we extensively review the available papers on IOT agreements, including the

ones with empirical estimations. Section (2.4) discusses the gap in the literature and how

this thesis can contribute to the literature.

2.1 International Roaming In EU

2.1.1 Sector Inquiry

In 1999, the EC opened a sector inquiry into mobile international roaming after receiving

numerous complaints. This section summarizes the information contained in the EC (2000)

document regarding the history of the IOT agreements and the initial findings.

Since 1992, members of the GSM Association (GSMA) applied the Normal Network Tariff

(NNT) regime for setting international roaming wholesale prices, which used the local retail

price for the equivalent service as a reference price plus a 15% markup. Because the

wholesale price was linked to the retail price in the visited market, it was subject to local

price competition. Visited networks kept changing the reference price to the more expensive

one in their range of retail prices (e.g. outside-the-package price for a minute of call).

In 1996, the GSMA notified its Standard International Roaming Agreement (STIRA) to the

EC for clearance from the cartel prohibition. STIRA clauses state that (i) wholesale service is

offered exclusively by licensed mobile networks (i.e., excluding virtual networks); (ii)

wholesale prices are non-discriminatory (i.e., applying same terms and conditions to all

members); and (iii) wholesale prices are uploaded to the GSMA Infocentre to be accessible

by a member network, where the member network cannot access the wholesale price set by

19

its competitor. The EC issued comfort letter conditional on the limited accessibility in item

(iii)11.

In 1998, the GSMA notified its IOT regime to the EC as the new regime for bilateral

wholesale price setting, replacing the NNT regime. The GSMA also received the EC’s

comfort letter. Since 1998, STIRA and IOT regime became the common framework for the

international roaming agreements.

The EC predicted that agreements based on the IOT regime would bring competition in the

wholesale market because domestic retail prices would not be used as reference prices.

Paradoxically, the referenced retail prices declined over time in response to domestic

competition, while international roaming wholesale prices increased dramatically.

For the sake of the sector inquiry, the EC defined the relevant national markets as: “The

market for retail roaming services, and the market for the provision of wholesale roaming to

foreign mobile network operators.” The initial findings in the EC (2000) are summarized as

follows.

Revenues from international roaming amount to 10-25% of the total revenue for a

mobile network.

The market has high barriers to entry with few network (oligopolists) that have similar

cost structures.

Wholesale roaming is offered exclusively by licensed public mobile networks, and

few, if any, are found to be dominant in the market.

Retail roaming is part of a bundle of mobile retail services, with almost complete

absence of retail competition. Other (outside) options, such as hotel phones or

international calling cards, are poor substitutes.

International roaming is a homogeneous product.

Due to lack of price transparency, consumers’ impact on pricing is trivial if they

override manually what their handsets had automatically selected.

Wholesale prices are set for each roaming service, which are usually based on

regional zones. Few networks offer wholesale discount, which is limited to the invoice

level (i.e., at the IOT bill level, rather than the listed wholesale price).

The retail price equals the wholesale price plus a handling charge (markup) of about

10-35%, which usually differs by visited networks.

11 See Valletti (2004) for possible remedies on STIRA clauses.

20

During the years 1997-2000, wholesale prices between EEA networks witnessed

absolute increases (126% for calling internationally and 166% for calling nationally),

with relative increases converging towards a higher overall price. All these price

changes bear no relation to the underlying costs. Consequently, retail prices

increased as well. These trends are in contrast to domestic prices.

Excessive pricing and tacit collusion (via identical pricing) are likely to be taking place

in the pricing behaviours of networks.

2.1.2 Aftermath Of Sector Inquiry

In this section, we summarize what happened after the EC’s 2000 sector inquiry till its

closure in 2007, relying on reports by Cullen International (2013).

The EC raided nine mobile operators in July 2001 in Germany, the Netherlands and the UK

to gather evidence on alleged price fixing. Later on, the focus of the investigation shifted

from collusion to abuse of dominant position, where the EC compared wholesale prices of

international roaming to domestic mobile termination rates since both share considerable

similarities in their actual costs, but have huge price differentials.

In July 2004, the EC sent statements of objection to O2 and Vodafone in the UK on abuse of

their dominant positions under Article 82 of the EC Treaty. Each network constituted a

separate market in the period 1997/1998 until the end of September 2003, during which,

each network enjoyed a dominant position in the provision of wholesale international

roaming on its own network. The abuse consists of unfair and excessive wholesale prices

charged to other European mobile networks.

In February 2005, the EC sent similar statements of objections to T-Mobile and Vodafone in

Germany. T-Mobile (since 1997) and Vodafone (since 2000) each enjoyed a dominant

position in the provision of wholesale international roaming on own network till the end of

2003.

In these statements of objections, the EC applied a market definition whereby each visited

network is considered a monopolist12. This definition is different to the one set out in the EC

2003 recommendation on relevant markets, which defines the market for international

roaming (known as Market 17) as: “Wholesale national market for international roaming on

12 Martino (2007) supports such market definition. Martino views if demand is randomly assigned between visited networks, then each network is a monopolist over its share of traffic. This randomness makes steering by the home network ineffective.

21

public mobile networks” (O.J., 2003). This definition includes all mobile networks in a given

country.

Under the EU 2003 regulatory framework13, international roaming was among 18 markets

that national regulatory authorities (NRAs) must review as part of market analysis procedure.

During the years 2005-2006, six NRAs finished their market analysis exercise for Market

(17). Their definitions of the geographic market corresponded to the EC’s definition of Market

(17). For analysis purposes, the NRAs considered outgoing calls as a product market, where

half of the NRAs calculated market shares in revenues and the other half market shares in

minutes of use (i.e. quantity).

The EC stated that the analysis of dominant position should be based on comparing market

shares with/out intra-groups (or cross-owned networks, e.g. Vodafone in the UK and in

Germany). NRAs which compared market shares with/out intra-group are Slovenia and Italy.

Table (2.1) summarizes their findings.

The six NRAs took into account expected competitive pressures on wholesale prices caused

by traffic steering as a countervailing buying power by home networks. All NRAs did not find

either single or joint significant market power. Their conclusions were based on the reduced

risk of coordinated behaviours due to market structure, growth of the market and its

seasonality nature. In addition, the existence of roaming alliance discounts were found to

reduce wholesale price transparency, and hence credible retaliation mechanism to

discourage deviation was assumed absent. In summary, all NRAs concluded that the market

was effectively competitive despite high roaming prices.

Although all NRAs found the market competitive at the wholesale level, the EC was

concerned that reductions in wholesale prices were not passed on to consumers. It was also

concerned that prices (retail and wholesale) remained unjustifiably high and showed no

signs of decline.

In June 2007, the EC issued roaming regulation to address the concerns of its sector inquiry

and law proceedings. The EC closed the sector inquiry, and closed the competition law

proceedings against the UK mobile networks (O2 and Vodafone) and against Germany

mobile networks (T-Mobile and Vodafone). In addition, the EC removed Market 17 from its

recommendation on relevant markets.

13 See Buigues and Rey (2004) for market definitions in the telecommunications industry.

22

1Table (2.1) NRAs’ responses to the EC market analysis notification exercise for Market (17).

Source: Adapted from Cullen International (2013).

Country

(Date of report) Notes

Finland (15 December 2005)

Market shares were calculated in quantities for 2005:

Sonera (45%) Elisa (39%) Finnet (14%) Alands (2%) Rise in traffics, steering and wholesale prices

Italy (7 June 2006)

Market shares were calculated in revenues for 2005:

TIM (38.8%; or 51.2% without intra-group) Vodafone (42.6%; or 23.9% without intra-group) Wind (18.3%; or 24.8% without intra-group) H3G (0.3%, or 0.1% without intra-group)

TIM’s share exceeded 50% when intra-group was excluded, but since its share by

volume was higher than share by revenue, TIM was conceived unable to charge higher wholesale price than its competitors

Denmark (28 July 2006)


A (37%) B (25%) C (38%)

High level of wholesale prices that did not change since 2001 Mobile operators belonged to different roaming alliances in Europe, making it

difficult to coordinate nationally

Slovenia (7 August 2006)


Mobitel (47%; or 55% without intra-group) Simobil (50%; or 42% without intra-group) WWI (3%; or 3% without intra-group)

High level of wholesale prices

Austria (25 August 2006)


Mobilkom (<40%) T-Mobile & Telering (merged) (>40%) The rest for One & Hutchison

70% of overall demand was from 5 countries, while 56% was from subsidiaries of

T-Mobile and Vodafone No network could charge higher wholesale price than its competitors, and all tried

to grant discounts to large buyers. Significant reductions in wholesale prices since 2001, particularly for alliance

networks

Spain (25 August 2006)


Vodafone (44%) Telefonica (39.6%) Amena (16.4%)

Home networks were able to steer 75-80% of traffics Difficulty in determining the actual wholesale price due to existence of discounts

23

2.1.3 EC Roaming Regulation

In June 2007, the EC roaming regulation applied price-caps on both wholesale and retail

prices, known as the Eurotariffs. The regulation also mandated home networks to send free

SMS to their travelling subscribers informing them about the retail prices14.

According to Falch et al. (2009), before the EU roaming regulation, retail prices for

international roaming calls were approximately four times higher than for national mobile

calls, indicating prices were exceeding underlying costs. Such excessive pricing provided

justification for the EU regulation, after a lengthy review from 1999 to 2007. In September

2007, the new regulation was fully implemented; as a result, retail prices were reduced by

57% for outgoing calls and 60% for incoming calls. After 2007, the EC amended the

regulation to lower the Eurotariffs and to include other services such as SMS and data

roaming.

2.2 Vertically Related Industries

The literature on vertically related industries addresses the vertical relationship of firms. It

investigates incentives for maximizing the individual profit versus joint profit and the choice

of contract (i.e. vertical restraints such as retail price maintenance or franchise fee compared

to linear pricing)15.

The main incentive for an upstream monopolist and downstream monopolist to vertically

integrate is to overcome the double marginalisation problem (Spengler, 1950). This problem

exists when the wholesaler marks up the input sold to the retailer, with the retailer itself also

adding another markup. Vertical integration internalizes this vertical externality by charging

the monopoly retail price, which increases producer surplus; and since this monopoly price is

lower than the double-markup price, consumer surplus is also increased. Therefore, total

welfare is enhanced. The incentive to vertically integrate may not hold under competition

and/or the use of vertical restraints.

Bonanno and Vickers (1988) develop a price game model for the case whereby a wholesaler

and a retailer compete with another pairing in the provision of final goods. The game is

14 This transparency measure was implemented in the Arab World. For a chronology of regional regulatory actions in this regard, see Sutherland (2008) for the EU case and Sutherland (2011) for the Arab World. 15 See Vickers and Waterson (1991) for an introduction on the economics of vertical integration and vertical restraints.

24

played in two stages, where the wholesaler sets a linear wholesale price (with/out a

franchise fee), and the retailer sets the retail price. If the goods are independent (monopoly),

the wholesaler’s profit from vertical integration is greater than the profit from vertical

separation due to elimination of double markups; but equal if franchise fee is used. On the

other hand, if goods are substitutes, the wholesaler can use the franchise fee to extract the

retailers’ surplus and subsequently charges a wholesale price above marginal cost. This

results in greater total profit under vertical separation than under vertical integration16.

The model in Economides (1994) is similar to the one in Bonanno and Vickers (1988),

except that Economides restricts the model to linear pricing and assumes that each good

consists of two complements in fixed proportions. These complement goods are produced

by two separate producers17. With linear demands for the final goods, the game is played in

two stages: producers decide whether or not to merge; then they set their input prices. The

game makes provision for four ownership structures: independent ownership; partial vertical

integration; parallel vertical integration; and joint ownership. The equilibria are then

compared at different degrees of substitutability. Playing the game out, Economides

concludes that when final goods are poor substitutes, producers have an incentive to

integrate; but this incentive diminishes as goods become perfect substitutes.

Telecommunications networks are similar to vertically related industries as they involve

wholesale pricing for handling interconnection services, such as terminating a call that is

originated on a different network. Networks would typically sign bilateral interconnection

agreements, commonly taking the form of two-way access agreements with linear wholesale

pricing. Examples include international settlement rates for international calling, domestic

access agreements for terminating local calls (e.g. mobile-to-mobile (MTM), mobile-to-fixed

(MTF) and fixed-to-mobile (FTM)), and IOT agreements for international roaming. In the

context of vertically related industries, the proceeding sections will review some papers on

international settlement rates18 that share similarity with IOT agreements, and following this

we review a paper on FTM that is referred to in the existing literature on international

roaming.

16 See Cyrenne (1994) for a closed form model in this line which only involves linear pricing. 17 That is, given fixed proportionality, the sum of the input prices equals the final price of the good. An example of this case is given in Section (2.2.2). 18 See Armstrong (2002) for a comprehensive summary on the theory on access, including international settlement rates, and see Einhorn (2002) for a literature survey on international settlement rates.

25

2.2.1 Access Theory

The similarity between IOT agreements and international settlement rates lies in the nature

of two-way access agreement between non-rivals that involves wholesale linear prices. The

literatures for both agreements share similarity in strategies and payoffs.

Consider the following simple example. If two non-rivals are monopolists within their

respective countries, assuming zero marginal costs for simplicity, the profit function to

partner is

( ) ( ) ( )⏞

( )⏞

( )

(2.1)

where , , and are retail price, wholesale price and demand respectively. Notice Eq. (2.1)

is separately additive: there exists profits derived from retail and wholesale. Each partner of

an agreement is a wholesaler and a retailer and therefore the wholesale demand for one

partner is the retail for the other. This typically represents the profit maximization problem for

international settlement rates, and IOT agreements as well.

Carter and Wright (1994) provide a model for two countries, each is served by a monopolist

offering international calling services and terminating international calls from the other

country (i.e. symbiotic producers). Their main finding is that, unilateral wholesale pricing

results in profits that are Pareto inefficient (due to double markups), and both networks can

be made better off by an appropriate choice of wholesale prices. Therefore, as the incentive

to remove the double marginalisation problem works both ways, both networks can

maximize their joint profits by accepting the division of profits with the sole use of wholesale

linear pricing, (while the retail price is set independently). This removes the need for other

mechanisms such as franchise fees.

The authors found that removing the double marginalization problem has a second order

loss in own profit but first order increase in the other’s profit. This is clear with firms having

equal demands19 , and would hold if the asymmetry is not excessive. If excessive, the

network with the larger wholesale demand will do better by not cooperating.

Cave and Donnelly (1996) focus on understanding the circumstances that may lead to

nonreciprocal bargained prices. Networks use the unilateral wholesale price as a threat point

19 If demands are identical, networks can agree on a reciprocal wholesale price equals marginal cost; hence the net payment is zero, retail prices are lower and monopoly profit is achieved as demonstrated in a closed-form model in (Box 5.1) in Laffont et al (2000).

26

for the Nash bargaining solution. Each network maximizes its own profit with respect to the

pair of wholesale prices, then both networks bargain over their solutions. The authors found

nonreciprocal wholesale prices as the outcome if profits are unequal when evaluated at

wholesale prices that equal marginal costs.

Wright (1999) provides a model to test the USA’s NRA claim that artificially high international

settlement rates led to high retail prices for American consumers, and that the deficit in net

payments subsidized foreign networks in low-income countries. He assumes a

representative consumer’s utility for international calling, with two countries where each is

served by a monopolist. The author then allows for retail competition. He assumes here a

common wholesale price equal to the weighted average cost of incoming calls, with weights

determined by income levels. Income disparity in this setting increases the common

wholesale price, and this wholesale price is reached by Nash bargaining solution.

In Wright (1999), starting with the threat of no agreement (zero profit), the network with the

high profit from any agreement (due to high demand or low actual costs) will gain more; thus

it is willing to share some of its profit with its partner, resulting in wholesale prices above

marginal cost.

With comparative statics, Wright concludes that competition has strong pressures on both

retail and wholesale prices. However, with reciprocal wholesale pricing and asymmetry

between networks, wholesale competition may not be effective. This is because networks

can inflate the wholesale price to keep retail prices artificially high.

Shy (2001) agrees with Wright. Shy assumes asymmetric unit demands for international

calling in two countries. The game is played in two stages: wholesale pricing followed by

retail. For simplicity, he assumes the networks agree on a mutual wholesale price equivalent

to the average of both unilaterally-set wholesale prices. Due to asymmetry in demands, the

network with high (retail) demand has a negative net payment with the low-demand network,

but still makes a positive profit. Shy then assumes perfect retail competition, where the retail

price equals the wholesale price (i.e. the perceived marginal cost). He concludes that, in

equilibrium, if each network has zero net payment and its profit equals its retail revenue,

networks can use the wholesale price as a means of collusion by artificially inflating it in

order to raise the retail price to the monopoly level.

Nonetheless, Shy views the supply-side alternatives, which can bypass high international

settlement rates, challenges such collusion. For example, the home network can rout the call

27

through international resellers, internet telephony, or through the cheapest terminating

network in the destined country20.

On local access, we review a paper by Gans and King (2000) who model FTM calling

originated from a single fixed network and terminated on a mobile network, where there are

more than one mobile network. They assume uniform retail prices for FTM calls because

either the consumer is unaware to which mobile network the mobile number belongs (due

probably to mobile number portability), or because the fixed network is unable to price

discriminate by mobile network (i.e. unable to set discriminatory retail prices by destined

mobile networks). Therefore, the fixed network’s consumer is likely to base his calling

patterns on the average retail price.

The model is built on a two-stage game: each mobile network sets a wholesale price, and

then the fixed network marks up the average wholesale price weighted by the market shares

of the mobile networks. The solution method is backward induction.

The authors found that wholesale price increases in the number of mobile networks, and is

negatively related to own market share. A wholesale price rise by one mobile network raises

the final price, which reduces total demand. Such price rise can represent a profit gain to the

undertaking network, while any loss in sales is shared with the rest of the mobile networks.

This effect is larger with smaller networks as the smaller network sets a higher wholesale

price. However, with linear demands, changes in wholesale prices (due to changes in market

shares) exactly offset each other and hence the retail price is independent from market

shares.

Gans and King (2000) concluded that competition between wholesalers does not exist

because any undercutting by a wholesaler does not raise its market share. As the number of

networks rises, the effect of horizontal separation increases, pushing up the wholesale price.

The authors’ model assumes a uniform retail price. The literature on retail on-net/off-net

pricing by domestic networks is abundant, where networks take into account the strategic

choice on access (or termination rates)21. In this case, a uniform retail price entails that the

network does not set any price differential between on-net and off-net calling. On the other

20 This requires liberalisation of international gateway. 21 See Peitz et al. (2004) for a literature survey on the theory of national access; and Hoernig and Valletti (2012) for a recent survey.

28

hand, a discriminatory retail price entails that the network sets price differential between on-

net and off-net calling22.

In international roaming, there is a noticeable trend in shifting from discriminatory retail

pricing to uniform, as noted by Ambjørnsen, et al. (2011). In fact, the EC (2000) found a few

mobile networks that offer uniform retail price; while the EC (2006) noted that often mobile

networks offer uniform retail prices.

As shall be demonstrated below, the existing models on international roaming are based on

assuming uniform retail price. Therefore, the conclusions of Gans and King (2000) are

similar to Salsas and Koboldt (2004), Lupi and Manenti (2009) and Ambjørnsen, et al.

(2011). All these papers assume the subscriber of a home network is unaware of the retail

prices per the interconnecting networks (wholesalers); and the consumer only cares about

the average retail price. This average price is uniform since it does not discriminate between

wholesalers.

We draw attention to the perfect complementarity feature in the aforementioned papers. We

argue that by assuming uniform retail price, wholesalers appear in the demand functions as

perfect complements, rather than substitutes. An explanation is provided below on the effect

of perfect complementarity on equilibria, which leads to results similar to the ones found by

Gans and King (2000) and by the published papers on international roaming.

2.2.2 Perfect Complementarity

Sonnenschein (1968) points out to the duality between the two Cournot’s theories of duopoly

(where two firms sell identical products) and complementary monopoly (where two products

are used in fixed proportions), where the difference lies in the interpretation placed on the

symbols 23 . Singh and Vives (1984) found that with differentiated products 24 : “Cournot

(Bertrand) competition with substitutes is the dual of Bertrand (Cournot) competition with

complements.”

22 In this case, the price for off-net calling is viewed as a departure of termination rate from the true marginal cost of call-termination (Hoernig and Valletti, 2012).

23 That is, for two firms A and B, the Cournot indirect demand is given by ( ); while the complementary monopoly direct demand is given by ( ). 24 Singh and Vives use a parameter in the demand that can be positive for substitutes or negative for complements.

29

In this section, we illustrate perfect complementarity in order to show how the number of

perfect complementary firms impacts upon equilibria. Assume firms sell products in fixed

proportions to one buyer, where the buyer in this case is the complementary monopoly.

Let the buyer’s demand for the products be given by

∑

(2.2)

where is the product price by firm , and is the number of complementary firms.

Assuming symmetric marginal costs (normalized to zero), each firm unilaterally maximizes

the profit

( )

( ∑

+

The first order condition for firm 1, for example, is

and its response function is given by

∑

(2.3)

Eq. (2.3) is downward sloping with respect to price charged by the rest of firms (i.e. response

functions are strategic substitutes). If the rest of firms respond symmetrically, Eq. (2.3) can

be expressed as

( )

( )

If firm responds symmetrically to firm 1, the (unilateral) price equilibrium is given by

(2.4)

Thus, the final price the buyer pays is

∑

(2.5)

and the equilibrium demand is

(2.6)

30

In the duopoly case, therefore, , which also equals . In this case, equals the

quantity equilibrium under Cournot duopoly competition, reflecting Sonnenschein’s duality.

This is due to the downward sloping in the respective response functions.

Table (2.2) shows equilibria for different values of . As rises: (1) own unilateral price

decreases, (2) total price to the buyer increases while its demand reduces, and (3) unilateral

profit to each firm declines. If all firms merge, the buyer will be charged the monopoly price

(½) and firms can share the monopoly profit (¼).

2Table (2.2) Illustration of perfect complementarity at different values of .

Number of

(complementary) firms

Unilateral price by

each firm

Final price charged to the buyer

Quantity demanded

by the buyer

Unilateral profit to

each firm

Merger profit to each firm

1 1/2 1/2 1/2 1/4 1/4

2 1/3 2/3 1/3 1/9 1/8

3 1/4 3/4 1/4 1/16 1/12

4 1/5 4/5 1/5 1/25 1/16

Figure (2.1) below demonstrates the prices at different values of . Clearly, as the number of

firms increases, the buyer pays less to each firm ( , the dotted-line), but pays more in total

( , the dashed-line). These prices equally depart from the monopoly price (½).

1Figure (2.1) Prices of perfect complementary firms at different values of n.

Each firm unilaterally charges a markup; thus double marginalisation takes place horizontally

between complementary firms. This horizontal separation externality can be internalised

through a merger. Nonetheless, competition does not exist because firms are complements.

0.00

0.10

0.20

0.30

0.40

0.50

0.60

0.70

0.80

0.90

1.00

1 2 3 4 5 6 7 8 9 10

Price

Number of firms

np*

p-monopoly

p*

31

This section demonstrated the effect of perfect complementarity on equilibria. Perfect

complementarity is a feature in the demands used in Gans and King (2000), Salsas and

Koboldt (2004), Lupi and Manenti (2009) and in Ambjørnsen, et al. (2011).

2.3 Literature On International Roaming

In this section, we review the available published and working papers that theoretically

modelled international roaming. We also review the available empirical papers.

2.3.1 Published Papers

The first published paper is Salsas and Koboldt (2004). They assume consumers put mobile

handsets on automatic selection mode because consumers only care about the average

retail price. The home network marks up the average wholesale price weighted by the

market shares of visited networks. The base model does not assume any supply-side

substitution (i.e. no steering), and thus the choice of visited networks is assumed to be

random.

The demand facing a home network is assumed to be linear (i.e. representative consumer,

with a substitutability parameter referring to downstream competition). Home networks set

linear retail prices and the consumer decides on a home network to subscribe to, based

merely on retail price for international roaming (ignoring switching costs). The retail price in

this case is non-differentiated by visited networks (i.e. uniform retail). Assuming two

countries, each country has two networks with symmetric coverage and costs. The price

game is played in two stages: wholesale price setting then retail price setting, solved by

backward induction.

Due to perfect complementarity of visited networks, the unilateral wholesale price in Salsas

and Koboldt (2004) is above the monopoly price. Under collusion, the equilibrium price

equals the monopoly level. These wholesale price equilibria do not change with retail

competition25, as retail competition only reduces retail markup.

Salsas and Koboldt conclude that wholesalers cannot enlarge their market shares via

lowering wholesale prices because any undercutting has less than a proportionate impact on

25 The wholesale profit changes in the retail markup, but the effect cancels out in the first order condition of the wholesale objective function.

32

the average wholesale price and consequently on the average retail price. They also found

the equilibrium wholesale price increases by the number of visited networks. Visited

networks on the other side benefit from retail competition because retail competition lowers

the retail markup and enhances demand.

The authors extend the base model by allowing default market shares to be asymmetric.

This leads to wholesale prices being negatively related to market shares. The authors found

that the visited network with low market share has an incentive to increase its wholesale

price. This is due to its smaller impact on the weighted average wholesale price. The base

model’s retail price does not change with the exception that it marks up the weighted

average wholesale price, instead of the simple average. As in Gans and King (2000), the

retail price is independent from wholesalers’ market shares.

Salsas and Koboldt also evaluate the model where only one pair of networks engages in a

vertical merger. They found the independent networks would set higher wholesale prices

compared to the merged entity. These differences lead to asymmetric retail prices (lower

with the merged entity). This difference shrinks with intense retail competition. They judged

that merger is unprofitable in this case because the consumer is unaware of prices. Then

they allowed the merged entity to improve price transparency. The fully informed consumer

will subscribe to the merged entity at home and will choose it (manually) in the visited

country. In this case, they judged that merger is profitable to the merged parties, which of

course comes at the expense of the independent parties (that did not merge). The degree of

profitability is stronger under more intense retail competition.

Finally, the authors explore the case when all home networks can steer to the cheapest

visited network, using complex expressions for market shares. Depending on the

effectiveness of steering, wholesale prices will approach marginal costs; and as a

consequence, retail prices will decline.

Lupi and Manenti (2009) study the impact of traffic steering on wholesale prices, building on

Salsas and Koboldt (2004), but ignoring retail competition. Similar to Salsas and Koboldt,

Lupi and Manenti conclude that wholesale and retail prices rise with the number of visited

networks. They relate this result to random traffics, which means each visited network acts

as a monopolist over its share of traffic; in other words, the home network deals with many

monopolists.

33

Lupi and Manenti then use a parameter to denote the success of steering, where networks

fully overlap in their respective coverages. The authors argue that wholesale prices can

reach marginal cost only if steering is perfect (i.e. fully successful). If steering is imperfect,

there will be no equilibrium in pure strategies, as explained below.

With imperfect steering, each visited network has a share of random traffic (not successfully

steered) as a source of monopoly power, and a share of the steered traffic that induces

wholesale competition. The home network will assign a higher market share to the visited

network that offers the lowest wholesale price. Asymmetry in market shares (due to steering)

causes asymmetry in wholesale prices; hence no equilibrium in pure strategies.

Lupi and Manenti conclude that, under steering, visited networks will offer menu wholesale

prices to home networks. They will offer a low price if the home network steers, a high price

if the home network steers away and a middle price if the home network does not steer to

any. This middle price equals the unilateral wholesale price, which is equivalent to the

weighted average of the low and the high prices, since market shares under steering act like

weights. However, the retail price is unaffected, similar to Salsas and Koboldt (2004).

Lupi and Manenti also assume a game for steering investment decision. They found

investment is the equilibrium in pure strategies. Because of no impact on retail prices, the

authors concluded that steering investment is a waste of resources.

Finally, the authors explore vertical merger, where traffic is assumed random (i.e. steering is

absent). The cooperative wholesale price between members is equal to marginal cost. A

matching game is then played out where each network simultaneously has to choose

whether to vertically merge with one IOT-agreement partner. They found non-cooperation is

the unique equilibrium of the game, because the non-cooperating pair can charge a higher

price without loosing market share.

A recent paper is provided by Ambjørnsen, et al. (2011), who develop a model based on

Salsas and Koboldt (2004), and Lupi and Manenti (2009). Ambjørnsen, et al. assume perfect

retail competition between two home networks and oligopoly visited networks. The game has

three stages: visited networks decide to invest in coverage improvement, and then

wholesale prices are set followed by retail prices. The game is solved by backward

induction. They assume uniform retail prices because the consumer considers only the

average retail price. They also assume steering by home network is absent, and that a

34

visited network can increase its probability of being selected by roamers if it improves its

coverage.

The model by Ambjørnsen, et al. uses linear demands, where a retailer faces weighted

average wholesale prices. Because of perfect retail competition, the retail price equals the

weighted average wholesale price. The model predicts wholesale price increases in the

number of firms (beyond the monopoly level). This is interpreted as each visited network

raises its wholesale price because it knows that its price does not affect the probability of it

being chosen.

The model states that wholesale prices are strategic substitutes if traffic is random, but if

manual network selection is assumed (hence no random traffic), wholesale prices can be

strategic complements.

Ambjørnsen, et al. highlight the noticeable price policy shift from discriminatory retail prices

(by visited networks) to uniform, which took the form of zone (or regional) pricing with some

mobile network operators. They also predict that when home networks set a uniform zonal

retail price, the weighted average wholesale price can rise further.

Ambjørnsen, et al. concluded that fierce upstream competition would increase both

wholesale and retail prices26. They also found the wholesale price is decreasing in own

market share, and concluded that investment to increase market share, such as coverage

improvement, is not profitable because a viable alternative is to reduce the wholesale price.

The authors extend the model to allow for traffic steering, using a parameter for success of

steering. Steering causes downward pressure on wholesale prices, and consequently, on

retail prices.

2.3.2 Working Papers

Roaming is also studied by Tsyganok (2008), who extends Salsas and Koboldt (2004) model

on pricing behaviours by considering non-linear and asymmetric demands, and varying the

number of networks. Her results are in line with Salsas and Koboldt’s.

26 Retail price equals wholesale price because of assuming perfect retail competition.

35

Bühler (2010) assumes two countries; each is served by two networks that compete for

subscribers downstream and for hosting visitors upstream. Home networks are differentiated

on a Hotelling line, and offer only international roaming service. They charge a two-part tariff

(i.e. a rental/fixed fee, and a uniform price per call) for their post-paid subscribers.

Subscribers can register with one network only. Assuming symmetric costs and demand

parameters, and reciprocal wholesale price setting, the game is as follows: networks set

wholesale prices, followed by the retail prices, with subscribers deciding on which home

network to join. The model also assumes wholesale roaming is sold non-cooperatively

through IOT agreements, or through alliances, where each member of an alliance commits

to buy exclusively from the alliance member (based on the agreed upon mutual wholesale

price); but each member can however sell to outsiders. This implicitly assumes perfect

control over network selection (i.e. perfect steering).

Bühler found that each home network sets the retail price of the call equal to the wholesale

price because of perfect retail competition, while the fixed fee is set to extract consumer

surplus. In equilibrium, wholesale price equals the marginal cost for non-alliance members.

However, if one alliance is formed, the wholesale price between the members will exceed

marginal cost. If two competing alliances are formed, due to strategic complementarity of

wholesale response functions, both alliances charge symmetric wholesale prices that

exceed prices under a single alliance. The author also found that networks endogenously

form alliances in a stage prior to setting wholesale prices, as each network benefits from the

higher wholesale price. Therefore, two competing alliances are formed in equilibrium.

The intuition behind excessive wholesale pricing (under alliance) in Bühler (2010) is the

following. Due to full steering commitment, alliance members have zero net payments since

they have symmetric demands and profits. Each network gets retail profit only from rental

(while the price of calling equals wholesale price). It can be determined therefore that a

wholesale price increase has two effects on the retail profit, which are: (1) a direct negative

effect on profit, and (2) a strategic effect on softening competition for subscribers by inducing

the downstream competitor (through strategic complementarity) to offer less attractive

contracts to its own subscribers. If the retailer does not join an alliance, the first effect

dominates the second; but by joining an alliance, the first effect is offset by gains at the

wholesale level (since raising the mutual wholesale price raises wholesale profits).

Bühler extends his base model to cases of imperfect steering. If steering is absent, the

selection between networks is random and the equilibrium wholesale price exceeds the one

36

in the base model. This result holds with or without alliances; thus, alliance formation has an

impact on wholesale prices only if steering is effective. In the continuous case (imperfect

steering), Bühler found that without alliances, no equilibrium in pure strategies exists. With

two competing alliances, the effectiveness of steering (i.e. its ability to manipulate market

shares) reduces the equilibrium wholesale price, but it is still above marginal cost. In

summary, subscribers are worse off under alliances, since retail price of calling increases

while the rental fee is unchanged.

Hualong and Shoulian (2009) develop a model similar to Salsas and Koboldt (2004). Under

different scenarios, they analyse the case whereby two networks form an alliance first and

then charge a lower wholesale price to each other. They concluded that lowered wholesale

prices do not improve payoffs, while lower retail prices reduce profits. Therefore, the alliance

members can improve their market shares either by steering, or choosing retail prices

(discriminated by visited networks) in order to let the consumer manually select an alliance

partner. The additional payoff alliance members make is a loss to the independent networks.

Furthermore, Shortall (2010) and Lacasa (2011) address the choice of alliance member

when demands are asymmetric. Both papers concluded that, using different scenarios and

numerical examples, networks with large retail demands will form an alliance, while small

networks will form another alliance.

2.3.3 Empirical Literature

This section surveys the available relevant empirical papers. In Section (2.1.2), we

summarized the analyses done by the European NRAs regarding steering and wholesale

pricing between European mobile network operators. The focus in the NRAs analyses was

about traffic steering as a countervailing buying power by home networks.

Most of the analyses compared a visited network’s market shares and wholesale prices over

two periods. The NRAs considered aggregate volumes (either traffics or sales), but only two

NRAs looked at these volumes with/out intra-group (e.g. networks that are cross-owned).

Using outgoing minutes as the product market, half of the analyses calculated market shares

based on revenues and the other half on minutes.

The NRAs reached a consensus that steering exists, which should enhance wholesale

competition. In Italy, for example, Vodafone has a market share of 42.6% in wholesale

revenues. By excluding intra-groups (revenues related to home networks that might have

37

preferential IOT agreements in Italy), Vodafone’s market share reduces to 23.9% (see Table

2.1). Therefore, steering by some of the intra-group home networks to Vodafone might be

behind the differences in market shares. For example, Vodafone in another country might be

steering its roamers in Italy towards Vodafone-Italy. However, this can be due to demand-

side substitution because of the existence of Vodafone’s Eurocall that offered lower retail

price when roaming on Vodafone networks.

Another data analysis topic in the EU is own price elasticity of demand, which has been

crucial in the debates regarding the impact assessment of the EC roaming regulation on

welfare. The EC (2006) used three scenarios for own price elasticities in its assessment. The

scenarios are (-0.55), (-1.0), and (-1.2). The GSMA (2008) disagrees with the EC, and

proposes that the elasticity should be (-0.25).

We found two industry studies that derive own price elasticity of demand. The first is a study

conducted by Europe Economics (2008) for the European Parliament. The second is a study

conducted by the NRA of Spain (CMT, 2009). The focus of both studies is estimating

demand elasticity for outgoing calls while roaming within the EU. The studies use roaming

total minutes of outgoing calls and average retail prices, all at the country-level, where the

study period is the second quarter of 2007 (i.e. a quarter before the EC roaming regulation)

to the first quarter of 2008, while CMT also covers up to the third quarter of 2008.

In Europe Economics (2008), the following model is estimated with panel fixed and random

effect regressions.

( ) ( )

(2.7)

where ( ) is the natural log of total minutes consumed in country in time ; ( ) is

the natural log of the average retail price per minute in country and time 27; is a

vector of time dummy variables to capture seasonality effects; and is the error term.

The coefficient in Eq. (2.7) represents the own price elasticity, which is estimated as (-

0.35) in the random–effect model, and (-0.44) in the fixed–effect model, where the

significance level is 5% in both estimations.

In CMT (2009), the following model is estimated with panel fixed-effect regression.

27 The use of lagged-time is due to assuming consumers require some time to adjust their behaviors in response to the price change.

38

( ) ( )

(2.8)

where ( ) is the natural log of total minutes consumed in country in time ; ( ) is the

natural log of the average retail price per minute in country and time ; is a vector of

control variables (price of SMS, number of tourists and summer-season dummy); and is

the error term.

CMT (2009) uses instrumental variables to solve the endogeneity problem of price and the

error term. The instruments are: the lagged price, a dummy variable if the country’s currency

is not Euro, and a variable to capture exchange rate variations for non-Euro countries.

The coefficient in Eq. (2.8) represents the own price elasticity, which is estimated as (-

0.37) in the model without instrument, and (-0.36) in the model with instrument, where the

significance level is 5% in both estimations.

Obviously, the aforementioned industry studies show that international roaming demand is

inelastic. This finding serves the argument against the EU regulation. The GSMA (2008)

argues that demand inelasticity implies that consumption would not increase largely by

capping retail prices and therefore the impact on producer surplus would raise questions

about the expected enhancement of total welfare28.

In surveying the literature, we found one empirical working paper on roaming alliances by

Rieck et al (2005). We also review an empirical paper by Wright (1999) on international

calling, since it shares some similarities with international roaming. In addition, we

summarize a working paper by Martino (2007) who provides guidance on how future

research should test for steering.

Rieck, et al. (2005) analyze empirically the implications of roaming alliance formations on

roaming retail prices. They observe the retail prices for outgoing calls (while roaming abroad)

to home countries, as listed on the websites of 88 mobile network operators in the OECD

countries and five in Asian countries29 , from September to mid-November of 2004. To

overcome the complexity in retail prices, the highest retail price for roaming on a given

visited network is considered. Network A1 in country A is considered to be favoring network

B1 among the other networks in visited country B if the lowest retail price A1 charges is for

28 The GSMA (2008) argues that the price cap creates a focal point for pricing, which harms competition. The GSMA proposes that transparency measures can enhance competition instead of price regulation. 29 Namely, Singapore, Hong Kong, Malaysia, Taiwan and Thailand.

39

roaming on B1’s network, assuming the rational consumer would manually select the visited

network with the cheapest retail price. A dummy variable was created for the network

membership in an existing roaming alliance (e.g. Vodafone group, Freemove, etc…).

Based on literature on social network analysis, the authors compute two indices: a network’s

centrality (how many favoring relations a network has), and networks’ clique (a subset of

networks forming a group). The control variables used in estimations are a network’s total

revenues and market shares in local subscribers, and other country-level variables such as

population and GDP.

The centrality index is regressed on the dummy for alliance membership and the control

variables. The roaming alliance dummy is found insignificant. The results from the control

variables suggest that networks in countries with high export-to-import ratio would tend to

have a high degree of centrality. In addition, a high degree of centrality is associated with

intermediate sized networks (in terms of revenues).

The dummy for alliance membership is regressed on the clique index to see whether an

existing alliance behaves like a strategic alliance (where members are expected to be

involved in favoring relations). The authors found that only Vodafone group members

constitute a strategic alliance (i.e. involving themselves in reciprocal favoring in terms of

retail prices) compared to the rest of alliances30.

Wright (1999) empirically tests his theoretical predictions on international settlement rates.

The author uses data on wholesale and retail prices, and volumes (quantities) of

international calls between the USA and 167 countries over 17 years. He found that retail

and wholesale prices have a strong positive correlation.

Wright investigates the determinants of wholesale prices. He found wholesale prices

increase with income disparity between the USA and the foreign countries, but the

magnitude decreases after controlling for competition, where competition generally reduces

wholesale prices. Marginal cost proxies fall over time, which influence wholesale prices as

follows: (1) wholesale price increases with distance between the USA and the country, (2)

increases with the country’s area, and (3) decreases with the country’s population (indicating

the existence of economies of scale).

30 We think this result is caused by Vodafone’s cross-ownership of mobile networks in multiple markets. Vodafone, as an example, offered a Eurocall plan, which was a common retail price for roaming on any Vodafone network in Europe (O.J., 2001). Vodafone is able to offer such retail plans relying on its presence in many markets. In contrast, another alliance, say Freemove, consists of different networks that may not be able to maintain preferential wholesale prices in each market.

40

Martino (2007) defines steering in international roaming. He argues that market definition

should be based on the success of traffic steering. If traffic steering is not possible, each

visited network should constitute a separate market.

Martino tries to answer the question of how to assess the existence of steering. He

recommends comparing the market share of a visited network in serving a home network to

the visited network’s market shares in serving other home networks.

Martino proposes that the calculation of market shares should be done for each service (e.g.

outgoing calls) in traffic (i.e. quantities) over a year to smooth out the seasonality effect. To

explore for steering, he recommends studying the movements in market shares over years.

Information on networks coverage, quality of services, technological developments, lifetime

of handsets, preferential agreements, roaming alliance relations, and merger/acquisition

events of visited networks should be included within an appropriate dataset.

2.4 Discussion

In this section, we identify how the next chapters complement the existing literature surveyed

in this chapter.

We develop a base model on pricing behaviours under IOT agreements. The base model

assumes linear (wholesale and retail) prices, and compares equilibria under different

cooperative and non-cooperative strategies (with different states of substitutability). It relies

on the literature on international settlement rates, in terms of using similar profit functions

and game setting. Such profit functions and game settings are also used by Salsas and

Koboldt (2004), Lupi and Manenti (2009) and Ambjørnsen, et al. (2011).

Moreover, our base model assumes international roaming is an additional service offered by

mobile networks. This, however, is contrary to Salsas and Koboldt (2004) and Bühler (2010)

who assume networks only offer international roaming and compete for subscribers.

In line with the existing published papers on roaming, our base model assumes a world of

two countries, each served by two networks that compete at the wholesale level but hold

monopoly positions at the retail level. The base model compares equilibria under

41

discriminatory retail policy (i.e. retail prices discriminated by visited networks) and uniform

policy. In other words, it explores different wholesale pricing strategies with and without the

uniform retail pricing constraint. With this uniform constraint, the results in the published

papers surveyed in Section (2.3.1) are reproduced. We demonstrate that uniform retail price

assumption makes wholesalers appear in the demand as complements rather than

substitutes, such that the price maximisation problems provide results that are inconsistent

with wholesale competition.

We also model two preferential IOT agreements under cross-ownership and roaming

alliance strategies, where the former results in asymmetric retail prices (assuming manual

network selection by the consumer), while the later is built on uniform retail price (assuming

automatic selection by consumer). We also compare the cross-ownership model’s results to

the case where retail price is uniform, as in Salsas and Koboldt (2004), and Lupi and

Manenti (2009).

We found that cooperating to form a cross-ownership pair is a dominant strategy to each

partner under discriminatory retail prices, since retail prices reflect preferential wholesale

prices. However, the opposite is true under uniform retail price: non-cooperation is the

dominant strategy to each partner; which is also found in Salsas and Koboldt (2004), and

Lupi and Manenti (2009).

We model roaming alliance based on assuming all home networks set uniform retail prices,

which equal the monopoly price given home networks hold monopoly positions downstream,

taken into consideration implications of net payments, as in Shy (2001). Steering requires

handsets to be on automatic mode, which is the assumed case under uniform retail price.

We found that investment in steering affects wholesale competition and alliance formations.

That is, with investment in steering, the home network can cause downward pressures on

wholesale prices. The equilibrium is such that all networks invest in steering, and in this

case, it is a waste of resources because profits are reduced by sunk costs. This intuition is

also found in Lupi and Manenti (2009), and in Ambjørnsen, et al (2011).

The success of steering is assumed to be dependent on the substitutability of demand,

which is different to the success rate used in Lupi and Manenti (2009). If steering is absent,

wholesale price equals the monopoly price. This reduces as substitutability increases. The

result is different from menu pricing found in Lupi and Manenti (2009), and from above-cost

pricing found in Bühler (2010).

42

We found IOT agreement partners have incentive to form alliances only if mutual steering is

insured. Thus, steering is a necessity for alliance formation, as predicted by Bühler (2010)

and Hualong and Shoulian (2009).

Investment in network coverage is addressed in Ambjørnsen, et al. (2011), but our model

takes coverage as given because visited networks typically set their respective coverages to

compete for home subscribers, rather than for visitors.

We also show that with non-zero net payments due to asymmetry in market sizes, each

network has an incentive to form an alliance with the largest network. This contradicts with

the conclusions of Shortall (2010) and Lacasa (2011) who argue that small networks prefer

to form an alliance.

With regards to empirical papers, as far as we know, there is no existing paper about

competition in wholesale roaming, structural demand estimation for visited networks, or

about the impact of preferred networks on wholesale prices.

We use a dataset on traffics (quantities) demanded by Etisalat outbound roamers and

wholesale charges per roaming service for each visited network. We make a distinction

between visited networks that are cross-owned by Etisalat and the ones that are considered

by Etisalat as alliance networks. This is to emphasize that a home network can steer to an

alliance network without necessarily having an ownership stake in it, a fact that is ignored in

the European NRAs analyses which mainly consider cross-ownership (intra-groups).

More importantly, the dataset has observations before and after Etisalat retail policy shift

from discriminatory pricing (i.e. per visited network) to uniform (i.e. common across visited

networks). Before the policy, retail prices were discriminatory and hence the choice between

visited networks is assumed to be manually selected by roamers (i.e. demand-side

substitution). After the policy, retail prices become uniform, so the choice is assumed to be

made by Etisalat through steering (i.e. supply-side substitution), controlling for visited

networks non-price characteristics and assuming handsets on automatic mode.

We draw attentions to the retail price policy importance in steering, which is ignored in the

European NRAs analyses and in Rieck et al (2005). A roamer can override his home

network’s steering by manually selecting a visited network with the cheaper retail price. Such

obstruction has to be removed for steering to exist. In order to steer, the home network

43

should make the roamer indifferent between visited networks price-wise, which is assumed

possible through setting identical retail prices (uniform). In our dataset, we make use of

Etisalat retail policy shift, which should act as a shock in the data through which we explore

for steering.

In our dataset, we only observe one side of the IOT agreements; that is, what Etisalat pays

out to visited networks (its wholesale costs), which is also the visited networks revenues.

Therefore, the reciprocity in prices that may exist in IOT agreements cannot be inferred from

the dataset, neither on the retail level as in Rieck et al (2005) nor on the wholesale level as

in Wright (1999) who studied international calling.

Nonetheless, the dataset is used to improve our understanding about wholesale competition

between visited networks. Therefore, we estimate demands for visited networks, estimate

the relationship of market structure and preferential IOT partners with wholesale prices. In

addition, we explore the effectiveness of steering by the home network, Etisalat.

We study outgoing calls while roaming abroad, similar to the European NRAs analyses, but

calculate market shares in quantities (as opposed to shares in sales) since steering is

related to market shares in quantities; thus we separate out the impact of wholesale pricing

behaviors.

Unlike Rieck et al (2005) who use retail prices to study alliance formation, we have data on

wholesale prices which are very relevant to preferential IOT (wholesale) agreements.

Wholesale prices charged to Etisalat can be compared between its IOT partners, controlling

for its alliance networks and cross-owned networks.

We found that, after the uniform policy, steering by Etisalat is effective. Wholesale discounts

are offered by Etisalat alliance networks, but not by its cross-owned networks. Etisalat

roamers are found to have higher marginal valuations for visited networks that allow prepaid

roaming and networks that are alliance to Etisalat, but less for networks cross-owned by

Etisalat.

In addition, the number of networks is found to significantly reduce wholesale prices,

reflecting substitutability of visited networks. Moreover, demand is found more elastic than

the industry figures in EC (2006), GSMA (2008), Europe Economics (2008) and CMT (2009).

44

Chapter (3). The Base Model

The exisiting theoritical models on mobile international roaming are based on networks

choosing wholesale prices with uniform retail pricing constraint; that is, the retail price is

common or non-discriminatory by visited networks, as in Salsas and Koboldt (2004), Lupi

and Manenti (2009), and Ambjørnsen, et al. (2011), which are surveyed in Section (2.3.1).

The rationale for assuming home networks applying a uniform retail is the fact that the

subscriber in many situations only cares about the average retail price even if prices differ

due to probably different wholesale prices. Such assumptions are used in Gans and King

(2000) who model the case a fixed network subscriber calling a mobile number without

knowing the terminating mobile network. Models on mobile international roaming also follow

these assumptions for the roamer who either originates a traffic (e.g. making an outgoing call)

or terminates a traffic (e.g. receiving an incoming call) when roaming on a foreign visited

network without knowing the applicable retail price for using it. Nonetheless, all these models

assume the subscriber knows the average retail price he eventually pays to his home

network.

However, as discussed in Chapter (2), this uniform retail price in the demand equation

makes networks complements rather than substitues, removing the wholesale competition

the existing papers aim at modelling. As a consequence, these papers share similar results:

the final wholesale price charged to the home network by non-cooperative upstream

networks is higher than the collusive price, which increases with the number of upstream

networks. Perfect complementairty explains how these results can exist, as demonestrated

in Section (2.2.2).

The assumption on retail pricing policy for mobile international roaming, whether uniform or

discriminatory, has implications on the roamer’s handset network selection and on wholesale

competition as we shall demonestrate in this chapter. The discriminatory retail pricing policy

is overlooked in the lietraure, so we aim in this chapter to compare equilibria under this

policy to the uniform policy used in the literature.

This chapter lays down the base model for mobile international roaming for the

discriminatory and the uniform retail pricing policies. Section (3.1) explains the model

background. Section (3.2) outlines the game. Section (3.3) shows game results and welfare

implications. Section (3.4) discusses game intuitions and its connection to next chapter.

45

3.1 Model Background

3.1.1 Demand Side

We base most of our assumptions on the findings in the EC sector inquiry (2000). We make

three key assumptions with regards to switching behaviours of the consumer of mobile

international roaming.

The first relates to the decision of registering at a home mobile network operator (hereafter

home network). The consumer has to be registered at a home network in order to be eligible

to roam abroad with his home mobile number. Given roaming is only a component of a

mobile bundle that is not offered on its own, it is plausible to assume that subscribers do not

decide to subscribe to a home network based on roaming reasons31.

In addition to the assumed negligible weight for international roaming in consumer’s decision

to choose a home network, we further assume high switching costs between home networks

that outweigh any domestic competitive offering for international roaming. The absence of

mobile number portability in some countries is an example, which makes the subscriber who

plans to switch away from the current home network unable to use his current mobile

number. If international roaming is mainly about using the same mobile number, such

convenience will not be enjoyed abroad in this case had the subscriber domestically

switched to another home network. Another similar example is a subscription contract with a

locked handset, under which the subscriber will not enjoy the convenience of using the same

handset abroad with a SIM card from a different home network. Based on all the above, we

assume a home network is a monopolist at the retail level.

The second switching assumption is that the outside options in the visited country are

inconvenient. Examples of such options include the use of a visitor-SIM, a VOIP

communications via WIFI, a satellite handset, a hotel telephone and a public payphone. Due

to the temporary nature of visits, we assume those outside options are highly inconvenient

as they require technical know-how to use for communication.

The third switching assumption relates to the roamer’s selection between different mobile

network operators in the visited country (hereafter visited networks). Visited networks are

31 Europe Economics (2008) claim most consumers place little weight on roaming prices in deciding which home network they subscribe to.

46

assumed to be differentiated in their geographical coverages. Other product differentiation

dimensions such as brand are ignored in the base model for simplicity32. The roamer is

assumed to manually select the visited network (through his handset) with the cheapest

retail price as long as its coverage allows. If the retail price is uniform, we assume the

roamer would consider leaving his handset on automatic mode.

We assume complete information33 and rational economic agents. Since visited networks

lack information on the heterogeneity of foreign visitors 34 , we use the representative

consumer approach to model the demands for visited networks.

3.1.2 Supply Side

The travelling subscriber is assumed to choose manually the visited network for his home

network based on the visited network’s coverage and on the retail price he eventually pays

to his home network. With uniform retail, he is assumed to leave his handset on default

automatic mode because he has no price reason to manually select between visited

networks; and thus the handset would be picked up any visited network (more likely the one

with the strongest network signal).

In the case of uniform retail price, in order for the home network to ensure the selected

visited network is a preferred visited network (which may offer a lower wholesale price), the

home network needs to acquire steering technology, such as SIM application toolkit or over

the air programming, to steer to the preferred network. Investment in such technology is

supposed to bring efficiency gains to the home network. Steering, as a supply-side

substitution, requires overlapping coverages of visited networks in order to be effective,

where its success is based on minimizing random traffics. We defer the choice on acquiring

steering technology to Chapter (4).

For the demand to be satisfied, visited networks must provide geographic coverages. Both

demand-side and supply-side substitutions (through steering) need overlapping coverages of

32 The estimated demand in Chapter (6) addresses visited networks’ characteristics, which are: (i) home network’s cross-owned networks that may bear the brand of home network, (ii) home network’s roaming alliance networks; and (iii) home network’s partners that allow prepaid roaming. Chapter (6) tests the significance of these characteristics in explaining the mean utility of Etisalat roamers, where the unobserved characteristics (by the econometrician), such as coverage of visited networks, enter the error term. 33 Mobile international roaming had been blamed for lack of price transparency. Regulators in the EU and Arab states overcame such problem by requesting home networks to send SMS to their travelling subscribers detailing the applicable roaming prices. 34 Visited networks recognise the home network of the visitors, and are not expected to know the details of the foreign visitors. This is why, as observed in the market place (e.g. Etisalat), visited networks do not price discriminate (in their wholesale pricing) between the visitors from the same home network (e.g. business/consumer; postpaid/prepaid, etc.).

47

visited networks; otherwise each network is an isolated monopolist in its own coverage area.

Thus the higher the overlap, the higher will be the substitutability.

Typically, the home network signs IOT agreement with all networks in a given country so that

its travelling subscriber can have the most possible network coverage in order to be able to

use the service continuously35. Therefore, visited networks are assumed to be imperfect

substitutes regarding coverage; hence preferential IOT agreements are not assumed to

result in exclusive IOT agreements.

Usually, a NRA obliges its licensed mobile network operators to meet the minimum network

coverage requirement. In the literature on choice of network coverage 36 , the networks

choose their coverages strategically to compete for domestic subscribers, where coverage is

considered a product quality parameter. As coverage increases product differentiation for

domestic subscribers, at the same time, it increases product substitutability for foreign

visitors. We do not assume the choice on coverage is made for the sake of hosting visitors;

hence it is exogenous to the agents in our model. This also means coverage fixed costs are

irrelevant too37.

In the base model, we use a substitutability parameter as an index to show that visited

networks’ coverages are overlapping where substitution between visited networks can be

made. The nature of mobile networks’ coverages does not make them perfect substitutes, so

the substitutability parameter is bounded.

Furthermore, we assume networks provide equal coverages; and have unconstrained

capacities, symmetric constant actual marginal costs (normalized to zero without loss of

generality) on both the retail and wholesale service provisioning, and no fixed costs as

networks use existing facilities to serve roamers38.

The focus of the model is on the pricing behaviour of networks, not their market structures,

so we assume a fixed number of players (i.e., mature market). More specifically, we assume

35 This is also noted by the French NRA (ARCEP, 2006). 36 See Valletti (1999) for coverage as a quality parameter, and see Valletti (2003) and Fabrizi and Wertlen (2008) for the choice on coverage and its impact on national roaming agreements. 37 Ambjørnsen, et al. (2011) assume a stage in the game, before the pricing stages, where visited networks choose to invest in coverage improvement, especially in airports, to compete in hosting visitors. 38 CAMEL technology allows roaming by prepaid subscribers, which must be installed by both the home and visited network. The home network instantly deducts from the balance of the prepaid roamer according to his usage. The technology is expensive to acquire, but typically, the home network charges higher retail for prepaid (European Parliament, 2006). According to Etisalat, visited networks do not price discriminate in wholesale prices for prepaid versus postpaid. CAMEL technology is assumed to affect the market size only; but is not assumed to affect the pricing behaviours of visited networks, which is the focus of all theoretical models. Therefore, costs of CAMEL technology are ignored in all the theoretical models.

48

a duopoly game. Networks are independent of one another (no cross-ownership39), and

maximize their roaming profits 40 independent from other existing interconnection

agreements, such as the international settlement rate agreement for international calls.

3.1.3 Pricing

Wholesale prices under IOT agreements are known to be set on a linear basis (i.e. per unit

of use)41 . For simpification and in line with the EC (2000), the model assumes linear

wholesale and retail prices. We start with discriminatory retail prices. Then we assume

uniform retail pricing in line with the existing literature. Home networks are assumed not to

coordinate their retail pricing strategy as the retail price is irrelevant in a typical IOT-

agreement42.

In line with the literature, a two-stage pricing game is played: wholesale price followed by

retail price. In different specification of the model, wholesale price setting may stem from

maximizing individual wholesale profit. If networks in the vertical line set the wholesale price

to maximize their joint profits, they may avoid the double marginalization problem43. Two

collusive pricing strategies are also explored44: domestic networks fixing the wholesale price

to maximize their national cartel profit45; and an international cartel aimed at maximizing the

market profit46. We compare the four wholesale strategies equilibria under discriminatory and

uniform retail price setting that aim at maximizing the retail profits of a home network across

its IOT agreements.

3.2 The Game

3.2.1 Players

Similar to Salsas and Koboldt (2004) and Lupi and Manenti (2009), we assume there are two

countries each is served by two networks. Each network is assumed to be a monopolist in

39 In Chapter (4), we model cross-ownership, which is a wholesale pricing strategy that results in preferential wholesale prices between the agreeing networks. 40 We assume a home network offers international roaming as an additional service to existing mobile package of services, and therefore will continue to offer it as long as it gains non-negative profit. 41 That is, wholesale prices do not involve two-part tariff at the retail level or franchise fee at the wholesale level. This may be due to demand uncertainty of roaming. 42 See Section (1.1) on IOT agreements. 43 Since each player is a wholesaler and a retailer (i.e. symbiotic producers), the incentive to maximize the vertical profit comes from both sides (Carter and Wright, 1994).

44 Collusive agreements are thought to be unofficial but coexist with the official wholesale (IOT) agreement. 45 This practice is a domestic price fixing against a foreign firm, which is illegal in some countries such as the EU member states. 46 The GSM Association is an example of a global umbrella that has a history in issuing roaming price guidelines (STIRA) and lobbying against price regulation (see Section 2.3.3).

49

providing the retail service to its travelling subscriber; and at the same time, since it can be

used by foreign visiting subscribers, it competes domestically for the wholesale service.

Denote countries by and . A pair of networks from the two countries sign a bilateral IOT

agreement denoted by ( ), where * +. Hence, there are four IOT agreements:

( ); ( ) ; ( ); and ( ) . Each agreement defines the wholesale prices

charged between the agreeing partners. Figure (3.1) depicts the relationship.

Country A

A1 A2

B1 B2

Country B

2Figure (3.1) International roaming in two countries each with two networks. -The double arrow represents a pair of two networks in an IOT agreement.

3.2.2 Demand Specification

We use linear demands as in Shubik and Levitan (1980). The main feature of this linear

demand is the independence of the substitutability parameter from the market size, which we

let vary to compare payoffs without affecting market size47.

Let us use the subscript as the network in question and the superscript as the foreign

network it deals with. The following equations are for network which applies similarly to

any network. The representative consumer who subscribes to home network has

preferences to roam on networks and given by the following quadratic utility

(

)

( )

(3.1)

where

is the aggregate quantity demanded by ’s subscriber for roaming

in country ; and

are the applicable retail prices; and (the market size) and

47 In contrast, the market size of the demand in Dixit (1979) is given by ( ), which is reduced by the substitutability parameter .

50

(slope of demand curve) are positive constants. The substitutability parameter is ,

which is also the inverse of product differentiation.

The roamer maximizes his utility for roaming in country by choosing and

(3.2)

The direct demands for roaming on networks and respectively are given by

[ .

/

]

and

[ .

/

]

(3.3)

The demands are downward sloping, showing the own price effect is greater than the cross

price effect (i.e. ). If the prices are uniform (i.e.

), the effect of

is removed from the demands, resulting in identical demands

, -

(3.4)

Figure (3.2) shows the direct demands as in Shubik and Levitan (1980). In this Figure,

consider firm 1 as visited network and firm 2 as ; while home network faces their

aggregated demands. The line is the analogy to Eq. (3.3); the line is the

analogy to Eq. (3.4); the line shows the aggregate demand (e.g. ); and the

line shows the demand for firm 1 when it drives out firm 2 from the market (i.e., firm 2 sells

nothing)48.

48 The kink in the demand curve insures non-negative demands. That is, there exists

at point such that because firm 2 is priced out of the market (Shubik and Levitan, 1980).

51

3Figure (3.2) The demand facing firm 1. Source: Figure (6.1) in Shubik and Levitan (1980).

In our case, for example, products are differentiated in their geographic coverages, not as a

response to host roamers, but as a response to local licensing obligations and to local

competition for subscribers (i.e. exogenous reasons). Visited networks use their existing

coverages to host roamers. Given our assumption on symmetric coverages, the roamer may

have equal quantities demanded. In the coverage overlapping areas, the roamer can

substitute between visited networks.

We let the substitutability parameter represent the overlapping coverage. If , the

roamer views visited networks as non-substitutes. This is equivalent to non-overlapping

coverage. As rises, the roamer views visited networks as more substitutable. This

corresponds to the increase in overlapping coverage. In the limit of , visited networks are

perfect substitutes, which means coverage overlap is perfect.

Define ( ) as the index for the degree of substitutability49, where , ). is

equivalent to the cross price elasticity divided by the absolute value of own price elasticity

when prices are symmetric. For example, take the following demand

0 .

/

1

The own price elasticity is given by

49 This index is also used in Economides (1994), and is similar to ( ) typically used with Dixit’s demand as in Singh and Vives (1984).

, -

[ .

/

]

(

* , -

52

.

/

and the cross elasticity is given by

When prices are symmetric in equilibrium for example, can be expressed as follows

| |

. /

If , the services provided by visited networks are independent (isolated monopolists);

and if , the services are becoming perfect substitutes. In our model, corresponds to

the overlapping of coverages of visited networks, under which the roamer50 can substitute.

3.2.3 Payoffs

The IOT agreement ( ) results in (

) profits to each partner consisting of a retail

part and a wholesale part, where * +. The following is a separately additive profit

function for network based on the ( ) IOT agreement

(

)

(

)

⏞

(

)

⏞

( )

The retail profit to comes from the situation ’s subscriber roams on ’s network.

charges its subscriber

and pays to . (In country , network is constrained by its

rival ( * + ) as long as the roamer can substitute and the wholesale prices

are reflected in ’s retail prices).

The wholesale profit to comes from the situation ’s travelling subscriber roams on ’s

network. ( is also constrained by its rival ( * + ) through the retail prices

sets for roaming in country as long as the roamer can substitute and the wholesale prices

are reflected in ’s retail prices). Notice that a network’s wholesale demand is also its IOT

agreement partner’s retail demand.

Let ( ) denote the market profit (i.e., the sum of all profits for all networks). The market

profit can be expressed horizontally, where and are the sum of profits in country

and respectively,

50 In Chapter (4), is used in modelling the steering game to represent home network’s supply-side substitution.

53

( ) ∑

⏞

∑

⏞

⏞

⏞

Observe above that the summations for

and have identical dimensions. Hence,

( ) can also be expressed vertically as follows, where ( ) is the sum of profits to the

vertically related partners in the ( ) IOT agreement.

( ) ∑ (

)

⏞

( )

⏞

( )

⏞

( )

⏞

( )

⏞

( )

Along with the base model’s assumptions, all the above payoffs result in symmetric price

solutions. In other words, these profit functions do not reflect preferential wholesale

agreements, which are deferred to Chapter (4).

3.2.4 Timing Of The Game

The game is played in two stages. In Stage I, networks set their wholesale prices

simultaneously. In Stage II, networks set their retail prices simultaneously. The subgame

perfect Nash equilibrium (SPNE) is solved by backward induction. The retail price solution of

the second stage is a function of the wholesale price(s), which is substituted for to solve for

first stage wholesale price equilibria. We call such a function the retail price mapping

function (RPMF) following Carter and Wright (1994).

For the sake of simplifying notations, we solve for the situation ’s subscriber roaming on

’s network as governed by the ( ) IOT agreement. We show the retail price solution

for network and the wholesale price solution for , where all derivations are provided in

Appendix (B). With symbiotic production, solutions apply similarly to any network.

3.2.5 Wholesale Pricing Strategies

We explore the profit maximizing strategies for wholesale price setting in the following four

types:

i. International cartel strategy;

ii. National cartel strategy;

iii. Narrow vertical strategy; and

iv. Unilateral strategy.

54

As we shall see next, the maximization problems for the above strategies lead to symmetric

wholesale prices for all networks. The unilateral strategy is a non-cooperative strategy,

whereby a network sets its wholesale price to maximize its own profit. The other strategies

are cooperative.

In the narrow vertical strategy, two vertically related networks set wholesale prices to

maximize their (narrow) joint profit ignoring implications on other revenues from their other

IOT agreements51.

The national cartel strategy is whereby two domestic networks maximize their joint profit by

fixing the wholesale price to a foreign network with which they have separate IOT

agreements. The international cartel strategy is about all players joining a coalition to set the

wholesale price between themselves in order to maximize the industry’s profit.

3.2.6 Retail Pricing Policies

Retail prices are not part of any IOT agreement, and thus networks are not supposed to

coordinate retail prices52. We consider uniform and discriminatory retail pricing policies which

produce different retail price mapping functions (RPMFs) to solve wholesale price equilibria.

To simplify the model, we assume all networks apply the same retail policy in each

wholesale pricing strategy.

First, we assume the retail price is differentiated by visited networks (i.e. discriminatory

policy). Then, we assume the retail price is uniform. Price equilibria are presented for each

retail pricing policy. Equilibria and results of the uniform policy are similar to the ones found

in Salsas and Koboldt (2004), Lupi and Manenti (2009), and Ambjørnsen, et al. (2011).

The demands under the uniform policy is given by Eq. (3.4), without the substitutability

parameter. The problem here is that the relevant RPMF, once substituted into the demand

equation to solve for wholesale prices, shows visited networks as perfect complements

instead of substitutes. As a consequence, response functions are downward sloping due to

perfect complementarity of visited networks.

51 Chapter (4) explores cross-ownership strategy, which is a wider wholesale strategy that maximizes the joint profit for the two vertically related networks and includes profit margins from their agreements with the other networks. Such strategy results in asymmetric wholesale price solutions. 52 Networks in an IOT agreement are licensed in two different countries and both offer retail services to two distinct groups of subscribers.

55

3.2.7 Stage II

Discriminatory Retail Policy

With discriminatory retail prices, each home network sets retail prices differentiated by

visited networks. The demands are similar to Eq. (3.3). Consider the following home network

’s retail maximization problem when its subscriber roams in country

(

)

(

) (

)

(

)

(

) (

)

(

)

(3.5.0)

’s retail prices are inside its retail profits (the first and the third terms, where the demands

are given by Eq. (3.3)). The RPMFs for roaming on both visited networks are derived by

solving

simultaneously, yielding the following response functions

( ) (

)

( )

and

( ) (

)

( )

(3.5.1)

By substituting symmetrically for and

in each equation of (3.5.1), we get the

following RPMFs

and

(3.6)

The RPMFs given in Eq. (3.6) are the monopoly markups53. The RPMF is a markup only on

the relevant wholesale price due to the following two effects.

Consider in the last term of

response function in Eq. (3.5.1). Substitute with its

RPMF from Eq. (3.6). One can see the impact of is removed by tightening the margin on

53 The RPMFs are similar to Eq. (6.15) in Shubik and Levitan (1980) if , which is ( ) , where is the actual marginal cost.

56

using ’s network. This is the primary effect. The secondary effect is due to the positive

sign of . It means a higher price for using allows to charge a higher price for using

. With the common (the market size), the effect of substitutability between ’s IOT

agreements is removed from each RPMF.

Uniform Retail Policy

In the absence of steering54 and with uniform retail, each home network is assumed to set

identical retail prices for its subscriber when roaming in the visited country, resulting in

identical demands. The demands are similar to Eq. (3.4).

Consider home network ’s retail maximization problem when its subscriber roams in

country . The average wholesale price facing is (

) . The following

retail maximization problem is for when its subscriber roams in country

( )

( ) ( )

( ) ( )

(3.7)

’s retail price is inside its retail profit (the first term, where the demands are given by Eq.

(3.4)). The first order condition (FOC) must satisfy

By solving for , the RPMF is given by

(3.8)

Note importantly that, as compared to Eq. (3.6), depends on both and

. It is also

equivalent to Eq. (3.6) if wholesale prices are symmetric. By substituting the RPMF given by

Eq. (3.8) to solve for wholesale prices, the demand for a visited network becomes

*

(

)

+

Clearly in above, wholesale prices have the same negative sign; making the visited

networks appear as perfect complements, where the nature of the relationship of visited

networks should be substitution. For example, to consume one minute of outgoing call, you

only need to be roaming on one visited network; put differently, you do not need to combine

half the minute from one network and the other half from the other network. The

54 Chapter (4) looks at a steering model with uniform retail price.

57

complementarity in this case is similar to the case of a monopoly’s demand for inputs from

two (perfect) complementary producers (i.e. Cournot complementary monopoly theory).

3.2.8 Stage I

The four wholesale pricing strategies are listed below. In each wholesale strategy, networks

are assumed to set their (monopoly) retail prices independently according to the relevant

RPMF of the retail policy (discriminatory or uniform).

In the following objective functions, the first order conditions (FOCs) are with respect to

(i.e. ’s wholesale price to ), where ’s retail price is given by the RPMFs of Eq. (3.6)

for the discriminatory policy and by the RPMF of Eq. (3.8) for the uniform policy.

3.2.8i International Cartel Wholesale Pricing Strategy

This strategy sets wholesale prices to maximize the international market profit given by

(

)

( )

(

)

(

)

(

)

(

)

(

)

(

)

(

)

(

)

(3.9)

Eq. (3.9) shows the market profit equals the retail revenues of all networks as wholesale

profits/costs cancel out. Note the relevant wholesale prices will be embedded in the relevant

RPMFs.

Under Discriminatory Retail Policy

Using the demands as in Eq. (3.3), and substituting Eq. (3.6) in Eq. (3.9), we find a

direct function of , and the FOC is

( )

By solving for , the international cartel’s best wholesale price is given by

(3.10)

58

where ( ) . Since , ) , the only symmetric wholesale equilibrium is when

. Thus the equilibrium wholesale solution is

then from Eq. (3.6),

(3.11)

With a wholesale price equals zero (the marginal cost), the retail price is at the monopoly

level.

Under Uniform Retail Policy

Using the demands as in Eq. (3.4), and substituting Eq. (3.8) in Eq. (3.9), we find a direct

function of , and the FOC is

( )


(3.12)

The only symmetric wholesale price solution is zero. Thus the equilibrium wholesale solution

is


(3.13)

The wholesale price equals zero (the marginal cost), and the retail price is at the monopoly

level.

3.2.8ii National Cartel Wholesale Pricing Strategy

This strategy sets wholesale prices to maximize the national profit. The maximization

problem for networks in country is

59

(

)

(

) (

)

(

)

(

) (

)

(

)

(

) (

)

(

)

(

) (

)

(

)

(3.14)




By solving for , the national cartel’s best wholesale price is given by

( )

(3.15)

where ( ). By substituting symmetrically for in Eq. (3.15), then substituting for

, we get


(3.16)

With a wholesale price at the monopoly level, the retail price reflects the double

marginalization problem.





(3.17)

60

By substituting symmetrically for in Eq. (3.17), the solution will be indeterminate. If we

let

, the wholesale price solution will be the monopoly price, similar to the

collusion case assumed in Salsas and Koboldt (2004) and Lupi and Manenti (2009).

Hence, the equilibrium wholesale price solution that maximizes the national market profit is


(3.18)

The wholesale price is fixed at the monopoly level resulting in a retail price with double

marginalisation problem.

3.2.8iii Narrow Vertical Wholesale Pricing Strategy

This strategy sets wholesale prices to maximize the vertical profits of the two IOT partners

within each separate IOT agreement. The maximization problem for the networks in the

( ) IOT agreement is

(

)

( )

(

)

(

)

(3.19)

Eq. (3.19) is equivalent to the retail revenues of and as their wholesale profits/costs

cancel out. It does not include their profits from their separate IOT agreements with and

55.




( )

By solving for , the narrow vertical strategy’s best wholesale price is given by

(

*

(3.20)

where ( ). By substituting symmetrically for in Eq. (3.20), we get

55 A wider vertical (cross-ownership) strategy that includes such cross profits is deferred to Chapter (4).

61

( )


( )

(3.21)

Since , ), is negative if . The negative wholesale price reduces the retail price

and effectively enhances the demand for the IOT partner. The retail price approaches zero

(the marginal cost) as .




( )

By solving for , the narrow vertical strategy’s best wholesale price is given by

(3.22)

The only symmetric wholesale price solution is zero. Thus the equilibrium wholesale solution

is


(3.23)

The wholesale price equals zero (the marginal cost), and the retail price is at the monopoly

level.

3.2.8iv Unilateral Wholesale Pricing Strategy

This strategy sets wholesale prices to maximize the own profit for each network. The

maximization problem for is

(

)

(

) (

)

(

)

(

) (

)

(

)

(3.24)

62




By solving for , the unilateral strategy’s best wholesale price is given by

( )

(3.25)

where ( ). By substituting symmetrically for in Eq. (3.25), then substituting for

, we get

( )


( )

( )

(3.26)

The substitutability parameter reduces both prices. As , the wholesale price

approaches zero (the marginal cost) and the retail price approaches the monopoly price.





(3.27)

By substituting symmetrically for in Eq. (3.27), we get


(3.28)

63

The price equilibria in Eq. (3.28) match the equilibria found in Salsas and Koboldt (2004) and

Lupi and Manenti (2009). With wholesale price already above the monopoly level, the double

marginalisation problem is prominent in the retail price.

3.3 Results

This section discusses the game results regarding the response functions, payoffs and

welfare implications.

3.3.1 Response Functions

Under a given retail pricing policy, the response functions have the same sign with all

wholesale strategies of the game. However, the response functions in the two retail pricing

policies (discriminatory and uniform) have opposite signs. This leads to the following

proposition.

Proposition (3.1). In all wholesale pricing strategies of the game, wholesale prices are

strategic complements under the discriminatory retail policy, and strategic substitutes under

the uniform retail policy.

Proof. The derivative with respect to the competitor’s wholesale price is positive in the

response functions under the discriminatory retail pricing policy given by Eq. (3.10), (3.15),

(3.20), and (3.25); hence strategic complements. The derivative with respect to the

competitor’s wholesale price is negative in the response functions under the uniform retail

pricing policy given by Eq. (3.12), (3.17), (3.22), and (3.27); hence strategic substitutes.

Visited networks are substitutes in nature, such that in a price game, the response functions

should be upward sloping (i.e., positively related or strategic complements). This is the case

under the discriminatory retail policy. On the other hand, when the response functions are

downward sloping (i.e., negatively related or strategic substitutes) in a price game, we

conclude that networks are made perfect complements. This is the case under the uniform

retail policy, as visited networks carry the same negative sign in the demand.

Salsas and Koboldt (2004), Lupi and Manenti (2009), and Ambjørnsen, et al. (2011) all

assume the roamer only cares about the average retail price, which in this case is similar to

64

the uniform retail policy in our model. With uniform retail, the roamer is assumed to leave his

handset on automatic mode; and given absence of steering, network selection becomes

random. Random network selection is attributed by Ambjørnsen, et al. (2011) to be causing

the wholesale pricing response functions to become strategic substitutes. Ambjørnsen, et al.

(2011) argue that the response functions can be, instead, strategic complements if roamers

manually select between networks, or if steering is effective by the home network.

Nevertheless, none of the aforementioned papers recognize that with uniform retail price, the

nature of visited networks turns to complements, instead of substitutes. Competition,

therefore, cannot be modelled between complementary networks.

Notice that with the national cartel wholesale strategy, Eq. (3.17) is downward sloping, but it

would lead to indeterminate solution. Instead, we imposed a common wholesale price that

led to a monopoly wholesale price, following the treatment made by Salsas and Koboldt

(2004), and in Lupi and Manenti (2009).

3.3.2 Payoffs Comparison

All price equilibria of the game have a convenient expression in separating out (the

price when demand falls to zero) from a fraction of one (except for the wholesale solutions

that equal zero). Let and represent standardised wholesale and retail prices

respectively, as shown in the Table (3.1). Assume market parameters ( ) are symmetric

for all networks.

3Table (3.1) Standardised wholesale and retail prices under discriminatory and uniform retail policies.

.

, ). ( ) , ).

Strategy Discriminatory retail policy Uniform retail policy

Wholesale Retail Wholesale Retail

International

cartel

National

cartel

Narrow

vertical

Unilateral

( )

65

Based on Table (3.1), the relation ( ) holds for any value of under any strategy:

is less than indicating wholesale price is not greater than retail price; and is less

than one reflecting positive demand. The equilibrium profit to network ( is any network)

from an IOT agreement is given by

( )( )

⏞

( )

⏞

( )

⏞

(3.29)

where and are standardised wholesale and retail prices respectively. Obviously, the

equilibrium profit equals the retail revenue because, in equilibrium, the wholesale cost and

wholesale profit cancel out. Since each network has two IOT agreements, each network will

get in equilibrium.

Notice that ( ) in Eq. (3.29) equals the own market share, , times (the price

when demand falls to zero). Both of these are the intercepts of the line in Figure (3.2).

With ( ), is a strictly concave function and symmetric around its maximum (i.e.

½, the monopoly price), as depicted in Figure (3.3).

Notice also that the limit exists for any solution that involves , which means is a

rational function of ; hence continuous at any value of (with negative derivatives with

respect to ). In addition, since changes in do not affect the market size , the curve in

Figure (3.3) is stable at any value of . Therefore, this curve can take all the solutions in

Table (3.1); where any that is invariant with will be a fixed point on the curve, and any

that varies with will move on the curve to the left.

( )

1/4

0 1/4 1/2 3/4 5/6 1

4Figure (3.3) Equilibrium profit to network in each IOT agreement. - Networks are assumed symmetric, and the expression ( ) is normalized to one, which is responsible for the curve’s height.

66

Therefore, payoffs can be easily compared under any strategy/policy in a comparative static

analysis that varies the substitutability parameter (or its index ), which are highlighted in

the following remarks.

Remark (3.1). Assume symmetric demand parameters and the equilibrium profit to a network

from an IOT agreement is given by Eq. (3.29), with price equilibria given by Table (3.1).

Therefore

i. Under the uniform retail policy, all profits are invariant with the substitutability index . The

international cartel and narrow vertical strategies yield the monopoly profit. The national cartel

strategy yields a higher profit than the unilateral strategy, but is lower than the monopoly

level.

ii. Under the discriminatory retail policy, the international cartel strategy yields the monopoly

profit that is invariant with the substitutability index . The national cartel strategy yields a

lower profit that is also invariant with . When , the narrow vertical strategy yields the

monopoly profit, while the unilateral strategy yields a profit equivalent to the national cartel

strategy. As rises, the profit of the unilateral strategy approaches the monopoly level while

the profit of the narrow vertical strategy approaches zero.

Proof. By Eq. (3.29), the equilibrium profit to network from an IOT agreement is

( ), with ( ) normalized to one since it is identical to all networks. From Table

(3.1), and given ( ), the monopoly retail price is ½, and the monopoly profit is ¼,

where all net payments are zero. If two solutions under two different strategies add up to

one, the payoffs from both strategies will be equal. With the uniform retail policy, there is no

effect of (or its index ). The international cartel and narrow vertical strategies yield the

monopoly profit since they charge the monopoly retail price. Retail price is higher with the

unilateral strategy compared to the national strategy; hence the national cartel strategy

yields a higher profit compared to the unilateral strategy ( compared to ), where

both profits are less than the monopoly profit. Under the discriminatory retail policy, the

international cartel strategy yields the monopoly profit because it charges the monopoly retail

price, and the national profit yields a lower profit ( ), where the payoffs from these two

strategies are invariant to . When (or ), the narrow vertical strategy yields the

monopoly profit, while the national cartel strategy yields a profit equivalent to the unilateral

strategy. As rises (or ), approaches zero under the narrow vertical strategy

leading to zero profit; while approaches the monopoly price under the unilateral strategy

67

leading to the monopoly profit. The profits of the narrow vertical and the unilateral strategies

are equal at (or ), since in both strategies add up to one.

3.3.3 Welfare Implications

We calculate the unweighted total surplus for a country as a proxy for the social welfare

under the strategies analysed in this chapter. Take as an example country which has two

networks and retailing the service for their subscribers who visit country , and at the

same time hosting roamers from country . The consumer surplus in country is given by

(3.30)

where the utility is as given by Eq. (3.1). Denote equilibrium utility for each network’s

subscriber in country by ( ). Let us use as in Section (3.3.2) to represent the

standardised price under any strategy. The consumer surplus in country is given by

( )

(3.31)

Notice the derivative of Eq. (3.31) with respect to is negative, which shows that price

reduces consumer surplus.

On the other hand, each network has two IOT agreements with the networks in country .

Thus the producer surplus in country is given by

(3.32)

Let be the equilibrium profit from an IOT agreement to any network in country ( ),

similar to Eq. (3.29). Therefore, we have

( )

(3.33)

Adding Eq. (3.31) to (3.33), the total surplus for country is given by

( )

(3.34)

As with the consumer surplus in Eq. (3.31), the total surplus in Eq. (3.34) is unambiguously

reduced by the standardised retail price, , where ( ). Price therefore is a

sufficient statistics to identify consumer and total surpluses. Any reduction in wholesale price

enhances consumer and total surpluses as long as this would reduce the retail price.

68

World surplus can be the sum of both countries’ total surpluses. However, as Valletti (2004)

puts it, national regulators may have distorted incentives towards regulation as each is

concerned with the surplus of its own consumers and own networks.

3.4 Discussion

The base model in this chapter aims at modelling pricing behaviours of mobile networks in

IOT agreements, comparing equilibria of four different wholesale pricing strategies under two

retail pricing policies (discriminatory and uniform). The model uses a demand system which

makes the substitutability parameter independent of market size. This substitutability

parameter is interpreted in terms of overlapping coverages, in which the roamer can

substitute between visited networks. The substitutability parameter is allowed to vary for

comparative static analyses.

Under the discriminatory retail policy, all response functions are upward sloping; hence

strategic complements. There are two effects on prices: double markups and Bertrand

competition. The unilateral retail price lies between the retail prices of the two cartel

strategies (national and international). When is zero, the relevant wholesale prices for the

unilateral and national cartel strategies are the same and thus their retail prices are

equivalent. This is the double-markup effect on the retail price. As rises, the unilateral

wholesale price reduces, approaching the marginal cost, driving the retail price towards the

monopoly price (which is also the retail price under the international cartel strategy). This

effect is due to wholesale Bertrand competition.

On the other side, the narrow vertical strategy has the Bertrand competition effect only.

When is zero, the wholesale price equals marginal cost; but as rises, the wholesale price

becomes negative, dragging down the retail price towards marginal cost to enhance the

demand for the IOT partner. Intuitively, mutual wholesale price subsidy is necessary to bring

down the retail price.

On the contrary, under the uniform retail policy, the relevant response functions are

downward sloping; hence strategic substitutes. The wholesale price equals the marginal cost

under the international cartel and the narrow vertical strategies, and is above marginal cost

under both the unilateral and the national cartel strategies. Notice the national cartel

(cooperative) wholesale price is lower than the unilateral (non-cooperative) price. Because

69

visited networks appear in the demand as perfect complements rather than substitutes, the

higher the number of visited networks, the higher is the wholesale price to the home

network. Those perfect complementary visited networks can overcome the effect of their

horizontal separation problem by charging the monopoly price and share the monopoly

profit.

Payoffs under a wholesale pricing strategy may differ depending on the retail pricing policy

being used. The retail price does not go below the monopoly level, except for the narrow

vertical strategy due to the negative wholesale price if . While impacts equilibria

under the discriminatory retail policy, it has no impact on the uniform retail policy because it

is removed from the demands. All equilibria can be compared on the same curve of Figure

(3.3) for any value of . When is zero, the position of is either on the right side of the

monopoly price (i.e. the double markups area), or at the monopoly price. As rises, only

equilibrium prices of the narrow vertical and unilateral strategies under the discriminatory

retail policy move on the curve towards the left. Changes in allow us to compare the

Pareto optimality of strategies. For instance, a shift in strategy from narrow vertical to

unilateral strategy will involve an increase in both wholesale and retail prices. Price

increases are noticed in Europe (EC, 2000).

In addition to the comparative static implications, the model shows that a price reduction

under any strategy will always enhance both domestic and global unweighted welfares.

However, due to the cross-border nature of the service, price regulation requires the

cooperation of relevant regulators to jointly control wholesale and retail prices (Hoernig and

Valletti, 2012), as in the case of the EC roaming regulation in 2007.

The base model can be extended to oligopoly markets by replacing the own market size

( ) in the profit function of Eq. (3.29) by ( ), where is the number of visited networks

in each country. This follows from the extension of the demands in Shubik and Levitan

(1980). The equilibrium profit from each IOT agreement would reduce by . Nonetheless, if

the equilibrium profit equals the retail revenue and if each network holds monopoly position

downstream, the total retail revenues to each network will be unchanged.

The base model compared equilibria under the discriminatory and uniform retail pricing

policies. The substitutability parameter is found to place a downward pressure on prices

under the discriminatory retail policy. In contrast, it has no impact on equilibria under the

uniform retail policy. Equilibria under the uniform retail policy without steering in this chapter

70

are similar to the ones found in the existing literature, which obviously produce equilibria that

are inconsistent with wholesale price competition.

The base model does not address asymmetric wholesale price solutions that may arise from

preferential IOT agreements. Chapter (4) extends the base model by addressing preferential

IOT agreements between cross-owned networks, and between roaming alliance networks.

Prior to roaming alliance game, an investment decision in steering technology is played by

home networks. Steering is assumed to be effective only if the roamer is in overlapping

coverage areas and the retail price is uniform such that the roamer’s handset is put on

automatic mode 56 . The home network that acquired steering technology can execute

steering (i.e. as a supply-substitution) between visited networks, seeking efficiency gains.

56 The existing literature assumes lack of retail price information about visited networks, but the roamer knows the average retail price he pays to his home network; and thus intuitively, the roamer needs not to manually select between visited networks (i.e. leave handset on automatic mode).

71

Chapter (4). Preferential Wholesale Agreements

In the base model of Chapter (3), all wholesale price equilibria from the Inter-Operator Tariffs

(IOT) agreements are symmetric. Symmetry of wholesale prices is a result of assuming

symmetric actual marginal costs with all networks. It is also a result of assuming compliance

in the cooperative strategies (international cartel, national cartel, and narrow vertical) that

perused identical maximization problems in each strategy, without involvement in

preferential agreements.

This chapter explores preferential IOT agreements that lead to asymmetry in wholesale

prices, in line with the theoretical papers surveyed in Section (2.3), which will be referred to

accordingly in this chapter. Similar to this literature, the focus of this chapter is on two

wholesale pricing strategies that involve preferential IOT agreements: cross-ownership in

Section (4.1); and roaming alliance in Section (4.2). The intuitions from these preferential

agreements are discussed in Section (4.3).

4.1 Cross-Ownership Model

Mobile network operators may involve in merger and acquisition across the borders of their

NRAs’ jurisdictions. Some cross-owned networks may carry the name of the parent

company (e.g. Etisalat-UAE and its subsidiary Etisalat-Egypt); or may operate under a

different brand (e.g. Etisalat-UAE and its subsidiary Mobily in Saudi Arabia). The first case is

typically a merger or a branch of the parent company; and the second is typically shares

acquired by the company(s). There are many mobile network operators that engage in

cross-border merger/acquisition with existing foreign networks or acquire a license to

compete with the existing networks, for example, AT&T, Vodafone, Orange, Zain, and MTN.

Cross-border merger/acquisition is a vertically related cross-ownership, since the engaged

networks are licensed by different NRAs to operate in different markets, and terminate cross-

border traffics. Nonetheless, foreign merger/acquisition may need to be notified to the

authority in charge of merger regulation, which approval decision is normally based on

whether or not the merger/acquisition raises competition concerns in the local market.

In this section, we assume a game whereby the engaged (foreign) networks choose the

wholesale prices under preferential IOT agreements. Similar to the base model, retail prices

72

are being set independently. The game is a wider vertical strategy, which is an extension of

the narrow vertical strategy presented in Section (3.2.8iii) of the base model, for it includes

the partners’ profit margins from their separate IOT agreements with outsiders.

Salsas and Koboldt (2004) address a vertical merger (or cross-ownership) by one pair of

networks assuming uniform retail pricing constraint and retail competition without steering

investment. They found merger is unprofitable in this case because the consumer is

unaware of prices. If the consumer is fully informed, he will subscribe to the merged entity at

home and will choose its subsidiary (manually) in the visited country; hence, merger is

profitable in this case, especially with more intense retail competition.

Lupi and Manenti (2009) also consider a similar model to Salsas and Koboldt (2004), but

ignored retail competition. They found non-cooperation is the equilibrium (i.e., a prisoner’s

dilemma case), because the non-cooperating pair can charge a higher price without loosing

market share.

In light of these two papers, we follow the uniform retail pricing constraint and compare its

results to the case without such constraint. In other words, we compare equilibria under the

uniform retail policy to equilibria under discriminatory retail policy. First, the backgrounds of

the discriminatory and uniform retail policies are explained. Then, the game is detailed for

each policy, where results are compared in the concluding remarks.

4.1.1 Model Background

Similar to the base model in Chapter (3), there are two networks in country ( and );

and two networks in country ( and ). Each network signs an IOT agreement with

each foreign network. Each network is assumed to be a monopolist at the retail level and a

duopolist at the wholesale level. Networks are assumed to have symmetric constant

marginal costs normalized to zero and no fixed costs. Retail and wholesale prices are

assumed to be linear57.

We use Shubik and Levitan (1980) demand for a visited network given by

[ .

/

] ( ) ( )

If the prices are uniform (i.e. ), the demands will be symmetric

57 Roaming packages (or bundles) applicable on preferred networks may unnecessarily complicate the model.

73

, -

All networks are assumed to have symmetric demand parameters ( ), with and

with the substitutability parameter .

The pricing game is played in two stages. In Stage I, networks decide between vertical

cross-ownership and unilateral strategies for their simultaneous setting of wholesale prices.

In Stage II, networks set their independent retail prices simultaneously. The subgame perfect

Nash equilibrium (SPNE) is solved by backward induction. As in the existing literature, the

focus of the game is on whether or not there exists an individual incentive for cross-

ownership within the framework of our pricing game.

We assume the NRA in each country does not approve a cross-ownership that involves both

its licensed networks and a foreign network, due to competition concerns. Thus, in our case,

each network decides to join at most one cross-ownership; in other words, there can be at

most two cross-ownership pairs.

The choice on a cross-ownership partner does not matter given networks are symmetric; in

other words, asymmetry in demand parameters may result in a preferred cross-ownership

partner. Given the vertical relationship and the two-way wholesale charging, we assume

cross-ownership once chosen is internally enforceable.

In line with the base model in Chapter (3), the subscript will refer to the network in question

and the superscript for its IOT partner. For the sake of simplifying notations, we show the

solutions for the case and engage in cross-ownership, while the other pair ( and

) either plays non-cooperatively or engages in a counter cross-ownership. The game is

detailed for discriminatory and uniform retail pricing policies respectively.

4.1.2 Discriminatory Retail Policy

Assume all networks apply discriminatory retail prices. sets its retail price(s) to maximize

its total profits from its two IOT agreements

(

)

(

) (

)

(

)

(

) (

)

(

)

(4.1)

74

As in the base model, ’s discriminatory retail price mapping functions (RPMFs) are given

by58

and

(4.2)

The joint profit for the cross-ownership ( ) is given by

( )

(

) (

) (

)

(

)

(

) (

) (

)

(

)

(4.3)

The first order conditions (FOCs) for Eq. (4.3) with respect to the wholesale prices charged

between and (the insiders) are given by

( )

( )

where and

are embedded in and

RPMFs respectively. Showing the

solution for , we have

( )

(

) *

(

)

+ (

) *

(

)

+

( )

(4.4)

where ( ). By further substituting Eqs. (4.2) in Eq. (4.4) and solving for , we

have

then from Eq. (4.2), we get the retail price solution

(4.5)

which is the monopoly price.

The wholesale price charged between the insiders equals (the actual) marginal cost,

regardless of the wholesale response function of the outsiders ( or ), leading to the

monopoly retail price for roaming on the insider’s network.

58 Similar to the base model, the second order condition is negative in all retail (discriminatory or uniform) and wholesale maximization problems.

75

The insiders’ FOCs for Eq. (4.3) with respect to their wholesale prices to the outsiders are

( )

( )

Showing the solution for , ’s response function is given by

( )

(4.6)

which is the baseline unilateral response function.

Eq. (4.5) and (4.6) will have different implications on price solutions depending on whether or

not the pair ( ) engage in a counter cross-ownership.

Lemma (4.1). With discriminatory retail policy of the model, if only one cross-ownership pair

exists and if , each of its members will receive independently a net payment from the

outsiders.

Proof. Consider the case only and engage in a cross-ownership. The insiders and

will set the internal wholesale price at marginal cost as per Eq. (4.5), and will use the

unilateral response function of Eq. (4.6) in choosing the wholesale prices to the outsiders

( and ). Since they do not engage in cross-ownership, and each uses a unilateral

response function similar to Eq. (4.6), by which they charge each other , ( )- (i.e.

the baseline unilateral wholesale price), and each charges , ( )- to the cross-

ownership members. In return, and will be charged the unilateral wholesale price. The

(retail) demands facing an outsider are evaluated at the unilateral retail price (

) , ( )-. For example, ’s demand for ( ) is

( ) , ( )-. By

contrast, the demands facing an insider are asymmetric because the insider charges

different retail prices: the monopoly price ( ) to use the insider’s network and (

) , ( )- to use the outsider’s network. For example, ’s demand for is

( ) , ( )-, while for , it is . The net payment between the insiders is

zero; and the net payment between the outsiders is also zero, because the relevant IOT

partners set identical wholesale prices and have identical demands. However, if , the

net payment between an outsider and an insider is non-zero: the outsider is a net payee to

the insider because the outsider charges a lower wholesale price (e.g.

) and has

a higher demand (e.g.

).

76

Lemma (4.1) shows asymmetry of prices and demands if one cross-ownership pair exists.

For example, the net payment for from its IOT agreement with is

( )

⏞

⏞

( )

⏞

( )

( )

⏞

( )

( )

where ( ).

Given discriminatory retail policy, the retail price for roaming on the insider’s network equals

the monopoly price, ( ) which is lower than ( ) , ( )- (the price for roaming

on the outsider’s network). As a consequence, the demands are asymmetric favouring the

insider’s network. On the other side, as the outsider is being charged identical unilateral

wholesale prices, it will set the unilateral retail price (i.e. ( ) , ( )-) for the use of

either visited network, resulting in equal demands. Each outsider has a higher demand for

the insider than vice versa, where the former charges a lower wholesale price than the later.

The result is a net payment from the outsider to the insider as long as networks are

imperfect substitutes (i.e., ).

Lemma (4.2). With discriminatory retail policy of the model, if two cross-ownership pairs

exist, net payments are zero.

Proof. Consider the case of two cross-ownership pairs, for example ( ) and ( ).

In each pair, the insiders will set the internal wholesale price at marginal cost as in Eq. (4.5);

and, with the unilateral response function of Eq. (4.6), they will set , ( )- as the

wholesale price to the outsiders. In each IOT agreement, wholesale prices are identical; and

demands are identical too (i.e. ( ) , ( )- for the insider, and for the

outsider). Therefore, net payments are zero.

Lemma (4.2) shows asymmetry in wholesale prices, but with zero net payments because

networks engage in two counter cross-ownership pairs. Notice that if , by moving from

the case in Lemma (4.1) to the case in Lemma (4.2), the wholesale price charged by and

to outsiders decreases (from , ( )- to , ( )-) and the retail price of and

for roaming on and respectively decreases (from ( ) , ( )-to (

) , ( )-). Lemma (4.1) and (4.2) lead to the following proposition.

Proposition (4.1). With discriminatory retail policy of the model, if , joining a cross-

ownership pair is a dominant strategy to each network.

77

Proof. Let represent the baseline unilateral profit to a network from its both IOT

agreements when no cross-ownership takes place in the market, all networks set wholesale

prices unilaterally, and net payments are zero; where ( )( ) , ( ) -.

From Lemma (4.1), the insider, for example , can improve upon its unilateral profit to

reach the cross-ownership profit ( ) , ( )- plus the net payment with

given by ( ) , ( ) - . By contrast, the outsider’s unilateral profit is

reduced by this net payment only if . From Lemma (4.2), each network gets and net

payments are zero. Therefore, joining cross-ownership is a dominant strategy to each

network.

The following example demonstrates the dominant strategy as predicted by Proposition

(4.1). Let ( ) be a possible cross-ownership pair and ( ) be a possible counter

pair. Consider the decision in joining different pairs by and . Assume . As given in

the proof of Proposition (4.1), let , and respectively represent the unilateral profit,

the cross-ownership profit and the net payment to these two IOT partners, where

, and and can be equal depending on values. Table (4.1) shows the payoffs

matrix.

4Table (4.1) Payoffs matrix for joining a cross-ownership pair under the discriminatory retail policy.

Possible cross-

ownership pairs:

( ) ( )

Join ( ) Don’t join

Join ( )

Don’t join

.

As can be seen from Table (4.1), the game has a unique equilibrium, which is joining a

cross-ownership pair. The non-zero net payment is behind the individual incentive to join a

cross-ownership. If one cross-ownership exists, its members can improve their payoffs

beyond the unilateral profit, but reduce the unilateral profits of the outsiders by the net

payments.

The question of Pareto efficiency of the game depends on values, where . But there

exists a critical , which is , at which ; below which, ; and above

which, . In the limit of , and equal the monopoly profit, ( ). At , the

substitutability index given by ( ) equals . Therefore, the equilibrium of

the cross-ownership game is Pareto efficient when substitutability is not very high. As

78

substitutability rises, the profit from the outsider decreases while the profit from the insider

increases. In contrast, the baseline unilateral profit rises with substitutability, overcoming the

cross-ownership profit when substitutability is too high.

4.1.3 Uniform Retail Policy

Assume all networks apply uniform retail prices without investing in steering technologies.

Here, the substitutability parameter is removed from the demands. sets its uniform retail

price to maximize its total profits from its two IOT agreements

( )

( ) ( )

( ) ( )

(4.7)

where

, and

.

’s uniform RPMF is given by

(4.8)

Rewriting Eq. (4.3) for the uniform retail policy, the joint profit for the cross-ownership

( ) is given by

( ) ( ) ( )

( ) ( ) ( )

( )

(4.9)

The FOCs for Eq. (4.9) with respect to the wholesale prices charged between the insiders

are given by

( )

( )

where and

are embedded in and RPMFs respectively. Showing the solution

for , we have

( )

( ) ( ) *

( )

+

(4.10)

By further substituting Eq. (4.8) in Eq. (4.10) and solving for , we have

(4.11)

79

The wholesale price charged between the insiders equals (the actual) marginal cost,

regardless of the wholesale response function of the outsiders ( or ).

The insider’s FOCs for Eq. (4.9) with respect to their wholesale prices to the outsiders are

( )

( )

Showing the solution for , ’s response function is given by

(4.12)

which is the baseline unilateral response function.

Eq. (4.11) and (4.12) will have different implications on price solutions depending on whether

or not the pair ( ) engage in a counter cross-ownership.

Lemma (4.3). With uniform retail policy of the model, if only one cross-ownership pair exists,

each of its members will pay independently a net payment to the outsiders.

Proof. Consider the case only and engage in a cross-ownership. The insiders and

will set the internal wholesale price at marginal cost as per Eq. (4.11), and will use the

unilateral response function of Eq. (4.12) in their wholesale price setting to the outsiders (

and ). Since they do not engage in cross-ownership, and each uses a unilateral

response function similar Eq. (4.12), by which they charge each other , - (i.e. the

unilateral wholesale price), but each charges to the cross-ownership members. In

return, and will be charged the unilateral wholesale price. The (retail) demands facing

an outsider are evaluated at the unilateral retail price , -. For example, ’s demand

for ( ) is

. By contrast, the demands facing an insider are evaluated at

the retail price , -. For example, ’s demand for ( ) is . The net

payment between the insiders is zero; and the net payment between the outsiders is also

zero, because the relevant IOT partners set identical wholesale prices and have identical

demands. However, the net payment between an outsider and an insider is non-zero: the

insider is a net payee to the outsider because the insider charges a lower wholesale price

(e.g.

) and has a higher demand (e.g.

).

Lemma (4.3) shows asymmetry of wholesale prices and demands if one cross-ownership

pair exists. For example, the net payment for from its IOT agreement with is positive,

80

⏞

⏞

⏞

⏞

Given uniform retail policy, each network will have a uniform retail, even if it engages in a

preferential IOT agreement. In other words, retail prices do not reflect preferential wholesale

prices, contrast to the case of the discriminatory retail policy. The retail price set by an

insider is lower than the one set by the outsider, leading to a higher demand facing the

insider compared to the outsider. In addition, given the insider charges a lower wholesale

price to the outsider than the other way, a net payment is paid from the insider to the

outsider.

Lemma (4.4). With uniform retail policy of the model, if two cross-ownership pairs exist, net

payments are zero.

Proof. Consider the case of two cross-ownership pairs, for example ( ) and ( ).

In each pair, the insiders will set the internal wholesale price at marginal cost as in Eq.

(4.11); and, with the unilateral response function similar to Eq. (4.12), they will set as

the wholesale price to the outsiders. In each IOT agreement, wholesale prices are identical,

and demands are identical too (i.e. ). Therefore, net payments are zero.

Lemma (4.4) shows asymmetry in wholesale prices, but with zero net payments because

networks engage in two counter cross-ownership pairs. Notice that by moving from the case

in Lemma (4.3) to the case in Lemma (4.4), the wholesale price charged by and to

outsiders increases (from to ) and the retail price of and decreases (from

to ). Lemmas (4.3) and (4.4) lead to the following proposition.

Proposition (4.2). With uniform retail policy of the model, considering the game of joining a

vertical cross-ownership, unilateral wholesale pricing is a dominant strategy to each network.

Proof. Let represent the baseline unilateral profit to a network from its both IOT

agreements when no cross-ownership takes place in the market, all networks set wholesale

prices unilaterally, and net payments are zero; where , -. From Lemma (4.3),

the insider’s profit from cross-ownership, for example , is , - minus the net

payment to given by , -. This is less than the unilateral profit. By contrast,

the outsider’s unilateral profit is increased by this net payment. From Lemma (4.4), net

payments are zero and each network gets . Therefore, unilateral wholesale pricing (i.e.

no-cross-ownership) is a dominant strategy to each network.

81


(4.2). Let ( ) be a possible cross-ownership pair and ( ) be a possible counter

pair. Consider the decision in joining different pairs by and . As given in the proof of

Proposition (4.2), let , and respectively represent the unilateral profit, the cross-

ownership profit and the net payment to these two IOT partners; where and

. Table (4.2) shows the payoffs matrix.

5Table (4.2) Payoffs matrix for joining a cross-ownership pair under the uniform retail policy.

Possible cross-

ownership pairs:

( ) ( )


Join ( )

Don’t join

.

As can be seen from Table (4.2), the game has a unique equilibrium, which is non-

cooperation (i.e. setting the wholesale price unilaterally). The non-zero net payment is

behind the individual incentive not to join a cross-ownership pair. If one cross-ownership

exists, the net payment improves the outsider’s profit if they do not form a counter cross-

ownership.

The equilibrium is Pareto inefficient as all networks could be made better off if they form two

cross-ownership pairs because the profit would be greater compared to baseline unilateral

profit (i.e. ); hence a prisoner’s dilemma situation. This result matches the result in

Lupi and Manenti (2009), and is contrary to the discriminatory retail policy in Section (4.1.2).

Notice with the uniform retail policy, the wholesale price solution ( ) charged between the

outsiders is the price at which the demand equals zero (or reservations price). This is due to

the perfect complementarity problem of visited networks in the demand equation: if the

insider charges zero, the outsider charges the reservation price. However, the average

wholesale price in this case is at the monopoly level (i.e. ( )).

82

4.1.4 Concluding Remarks

In this chapter, the two models for cross-ownership, one with discriminatory retail policy and

the other with uniform policy, consider a vertically related IOT partners choosing wholesale

prices cooperatively, while retail prices are set independently. The discriminatory policy is

overlooked in the existing literature, while the uniform is analysed in Salsas and Koboldt

(2004) and Lupi and Manenti (2009), where steering is ignored to maintain simplicity.

As can be seen in Table (4.3) below, wholesale price solutions are found asymmetric under

both retail policies: the cross-ownership members set the internal wholesale price at

marginal cost and the external wholesale price is set above marginal cost. In the case of one

cross-ownership pair, prices and demands are asymmetric with the outsiders, resulting in net

payments in favour of the cross-ownership pair under the discriminatory policy, but in favour

of the outsiders under the uniform policy.

6Table (4.3) Standardised prices under discriminatory and uniform retail policies in the cross-ownership game.

.

, ).

Based on the findings of the models, joining a cross-ownership pair is a dominant strategy to

each network under the discriminatory policy59. On the other hand, joining a cross-ownership

pair is not a dominant strategy under the uniform policy, as also predicted by Lupi and

Manenti (2009).

59 Without detailed modelling, this result is referred to by Salsas and Koboldt (2004).

Scenario

Pricing with

each

partner

Discriminatory retail policy Uniform retail policy

Wholesale cost Retail price Wholesale cost Retail price

One cross-

ownership

pair

The joining

network

With insider

With outsider

( )

The non-

joining

network

With insider

( )

With outsider

( )

Two cross-

ownership

pairs

Each

network

With insider

With outsider

( )

83

The equilibrium under the discriminatory policy is Pareto efficient if the substitutability level is

not too high. With the uniform policy, the equilibrium is inefficient as networks can be made

better off if they form two cross-ownership pairs. This is because with cross-ownership, the

average wholesale price is reduced, eliminating some of the double-markup effects.

The discriminatory retail pricing policy relies on the consumer choosing (manually through

his handset) the visited network with the cheapest retail price. By contrast, the uniform retail

policy applies a uniform retail price. Given absence of steering and consumer’s indifference

when retail prices are identical (i.e. handset on automatic mode), neither the home network

nor the consumer would change the default choice on visited networks, despite the

existence of a preferred visited network which charges a lower wholesale price. Therefore,

uniform retail is ineffective without steering. Next, we assume uniform retail is the main

assumption for the steering model, but we need to explore the impact of market shares

asymmetry on results under the uniform retail policy in the next section.

84

4.2 Uniform Price With Asymmetric Market Shares

This section explores the impact of market share asymmetry on wholesale price solutions

when the retail price is uniform. Uniform retail price is assumed necessary for steering to

exist, otherwise manual network selection by the roamer overrides steering made by home

network. We follow the steps of Salsas and Koboldt (2004), and Lupi and Manenti (2009)

who use asymmetric market shares in their modelling of steering, and explain how

asymmetry in shares does not impact the final wholesale price charged to the home network,

nor the uniform price it sets, which undermines the need for steering.

This section follows the wholesale unilateral pricing strategy assuming uniform retail policy

(as in the Sections 3.2.7 and 3.2.8iv) except that we replace the simple average wholesale

price with a weighted average wholesale price, where the weights are the market shares of

visited networks.

Let us consider the retail maximization problem for when its subscriber roams in country

, as in Eq. (3.7), but replace the simple average wholesale price with this weighted

average (

), where

and are the market shares of and

respectively in hosting ’s subscriber. By solving for , the RPMF becomes

(4.13)

The above RPMF is a markup on the weighted average wholesale price. The unilateral

wholesale price strategy sets wholesale prices to maximize the unilateral wholesale profit,

leading to, as an example, the following unilateral response function for

(4.14)

Notice Eq. (4.14) equals the baseline unilateral response function if market shares are

symmetric (i.e. ( )

). By substituting symmetrically for in Eq. (4.14),

we get these solutions

and

(4.15)

85

By taking the weighted average of the solutions in Eq. (4.15), using their market shares as

weights, we have

(4.16)

Obviously, the weighted average wholesale price in Eq. (4.16) equals the baseline solution

of the unilateral wholesale price. It is clear from Eq. (4.15) that the network’s own market

share reduces its wholesale price. However, the asymmetry in market shares does not

impact the average wholesale price (whether weighted average or simple) charged to the

home network. In this case, market shares asymmetry does not change the fact that visited

networks appear in the demand as perfect complements. Therefore, the home network’s

retail price is unchanged.

The above results match the ones found in Salsas and Koboldt (2004), Lupi and Manenti

(2009), and Ambjørnsen, et al. (2011); and are also similar to the results found in Gans and

King (2000) who modelled fixed-to-mobile access. As noted in Gans and King (2000), with

linear demands, the changes in wholesale prices due to changes in market shares exactly

offset each other. Therefore, the (uniform) retail price is independent from market shares.

The main problem in this setting, where retail price is uniform and market shares can be

asymmetric, is that if the home network is able to steer (and effectively manipulates market

shares of visited networks), it does not cause downward pressures on wholesale prices. This

undermines the efficiency gains steering investment is assumed to bring.

Another problem in using the uniform retail price in this setting is that the substitutability

parameter is removed from the demands. As substitutability does not exist in all relevant

solutions, the expressions of market shares do not reflect substitutability. We assumed

visited networks are substitutable if their coverages overlap, where steering is possible only

in the overlapping areas. In the next section on steering model, the expressions of market

shares take into consideration substitutability between visited networks.

86

4.3 Roaming Alliance Model

Traffic steering techniques allow the home network to manipulate the market shares of its

IOT-agreement partners in the visited country. This practice was deemed in the European

NRAs analyses, before the EC roaming regulation, as a countervailing buyer power to pay

cheaper wholesale prices. Moreover, the nature of the two-way access of IOT agreements

enabled the emergence of two-way steering, (hereafter roaming alliance).

In order for steering to exist, three conditions we assume are necessary: a uniform retail

price to avoid roamer’s overriding selection (i.e. handsets on automatic mode to allow

steering); expected efficiency gains (if wholesale prices are high to justify steering

investment); and of course overlapping coverages (i.e., substitutability between visited

networks).

Salsas and Koboldt (2004) use complex expressions for market shares, exploring the case

when all networks steer to the cheapest visited network. Depending on the ability to steer,

the authors think wholesale prices will approach marginal costs, dragging down retail prices.

Lupi and Manenti (2009) offer a detailed model for steering. They assume full overlapping

coverage, and use a rate of steering success. They concluded that with perfect success rate,

wholesale price is driven to marginal cost; if imperfect, there is no equilibrium in pure

strategies. That is, there will be asymmetry in wholesale prices: a low price to the home

network that commits to steer, and a high price if it steers away. The baseline unilateral price

is between these two prices, which is charged to the home network that does not steer to

any (i.e. does not acquire steering technology). As a consequence, each visited network

offers a menu of wholesale prices: a high and low price, and the unilateral price. In addition,

they found steering investment is the equilibrium in pure strategies; which is a waste of

resources since it does not impact retail prices that bear the double marginalisation problem.

However, the aforementioned papers rely on uniform retail price in the demand; thus, the

weighted average wholesale price to a home network equals the baseline unilateral

wholesale price, as demonstrated in Section (4.2). Therefore, retail prices and demands are

unaltered, given visited networks appear as perfect complements in this case.

87

Next section addresses our model backgrounds in the context of the existing literature. Then

two games are detailed: a network’s choice on steering investment; and a network’s choice

in joining a roaming alliance. Results are discussed finally.

4.3.1 Model Background

The existing steering models relied on the uniform retail policy. Yet this policy has four

problems. First, as explained previously, the uniform policy is inferior to the discriminatory

policy in the non-cooperative wholesale price setting. That is, the uniform policy is not

optimal for a retailer network with monopoly position. Second, the demand under this policy

makes visited networks appear as perfect complements rather than substitutes, which

removes upstream competition by removing the substitutability parameter necessary for

effective steering. Third, as explained in Section (4.2) and found in Lupi and Manenti (2009),

there will be no wholesale price equilibrium in pure strategies since changes in market

shares create asymmetric solutions. Fourth, complication of the steering model through

market share asymmetry in a two-stage pricing game is unnecessary since the retail price is

independent from market shares asymmetry, as also explained in Section (4.2).

We avoid the uniform retail policy assumed in the existing literature because of the above-

mentioned problems. Rather, we skip the retail maximization problem in the steering game,

and assume instead each home network chooses uniform retail price. It does not matter

which uniform price is chosen in this game as long as the status quo net payments are zero.

However, if net payments are zero such that the profit equals the retail revenue and given

unconstrained retailers, the optimal retail price is the monopoly price. In this case, the

unconstrained monopolist gains the monopoly retail revenue. Such monopoly price is

uniform by definition as it is common for the use of any visited network.

We maintain the same assumptions as in Section (4.1.1), unless mentioned otherwise. The

retail stage, Stage II, is skipped in this game by assuming all networks set the monopoly

retail price. For simplicity, the game is split into two static games:

i. A two-stage game, where each network at the retail level decides simultaneously on

steering investment, followed by simultaneous wholesale price setting; and

ii. A game whereby each network decides simultaneously on joining a roaming alliance,

where an alliance can take at most two vertically-related networks.

88

Note that, along with symmetry in demand parameters and in the absence of steering

investment, the assumption of monopoly retail price provides no incentive for wholesalers to

undercut because this does not raise own market share. In this situation, wholesale price is

fixed at the monopoly level and net payments are zero. Additionally, as explained in

Appendix (C), the assumption on monopoly retail price is in accordance with an infinitely

repeated game equilibrium. As a consequence, steering investment is deemed to be feasible

in this situation because efficiency gains are possible given wholesale prices are at the

monopoly level.

Networks are assumed to incur a common sunk cost, , for buying the steering technology.

Undercutting is given by the unilateral response function (e.g. , ( )-

),

where the baseline unilateral wholesale price is , ( )- and ( ) , the

substitutability index. Therefore, we can maintain the substitutability parameter, , indicating

coverage overlapping where steering can be exercised. Additionally, we can insure that

market share manipulation does not affect unilateral wholesale price equilibria.

In addition, we assume the home network’s success of steering depends only on coverage

overlap. Accordingly, there are no random traffics since the level of randomness that exists

in overlapping areas is successfully eliminated via steering. Let the visited network that is

being steered to receives the market share ( ) , and the remaining ( ) goes to

the visited network that is being steered away, where ( )60.

Coverages are assumed symmetric 61 . Steering between two visited networks implies

preferring only one network, where full steering can only exist in the limit of . Nonetheless,

given networks are assumed imperfect substitutes, an alliance formed for instance between

and can coexist with the existing roaming agreements with the other partners, and

. Furthermore, the choice on an alliance partner does not matter since networks are

symmetric. In other words, asymmetry in demand parameters may result in a preferred

alliance partner.

Based on the model backgrounds, in the status quo, both the retail and wholesale prices

equal the monopoly price, ( ). Accordingly, the monopoly retail revenues are unaffected

by market shares manipulation, because market shares add up to one. Moreover, the

60 These market share expressions are derived from Shubik and Levitan (1980) demand: if firm 1 drives firm 2 out from the

market, the demand to firm 1 becomes (( ) )) , -, where ( ) ( ) ( ) and ( ) , representing the line ( ) in Figure (3.2). 61 Contrary to Ambjørnsen, et al. (2011) who model investment decision to expand coverage in airports for example, we maintain the base model’s assumption that networks installed their coverages competing for home subscribers only, not for hosting visitors.

89

demand for a visited network is given by its market share times . We only show net

payments as payoffs to simplify presentations.

4.3.2 Steering Investment Game

Assume all home networks set the retail price at the monopoly level and the status quo

wholesale prices equal the monopoly price. Assume a two-stage game. In Stage I, home

networks decide simultaneously on investing in steering technology, which costs . In

Stage II, wholesale prices are set non-cooperatively. The solution method for the Subgame

Perfect Nash Equilibrium (SPNE) is backward induction.

Proposition (4.3). In the steering investment game, conditional on , ( ) -, the

SPNE of the game is such that each network invests in steering and sets wholesale prices

unilaterally.

Proof. The status quo is when steering investment does not exist, where visited networks

sustain the wholesale price at the monopoly level (i.e. , -), because undercutting is

unattractive as it does not raise market shares. When a home network invests in steering,

depending on the substitutability level, it can increase the market share of the cheapest

visited network to ( ) , while the other network gets ( ) , where ( ). In

this case, each visited network charges the baseline unilateral wholesale price (i.e.

, ( )-), which is the dominant strategy in Stage II. In Stage I, the home network gains

from the net payment with each visited network that does not invest in steering due to

asymmetry in wholesale prices (i.e., monopoly price by the former compared to baseline

unilateral price by the later). The expected gain makes investment a dominant strategy in

Stage I, as long as the fixed cost of investment , ( ) -.

Proposition (4.3) can be demonstrated as follows. Let denotes the net payment from an

IOT agreement, with subscript refers to the network in question and the superscript to its

partner. If invests in steering, and will set the baseline unilateral price (

, ( )-). If does not invest, will set the monopoly price ( , -); hence

will get a positive net payment as follows

( )

The incurred fixed cost must be less than or equal to the expected gain from each IOT

agreement. It is clear from the above example that the gain from wholesale price differentials

90

requires positive , which allows steering to be effective since visited networks are

substitutable.

The above net payment gain increases by (i.e. derivative of with respect to is

positive). With , net payments are zero since the baseline unilateral wholesale

price, , equals the monopoly price, .

Predictably, investing in steering causes downward pressures on wholesale prices in the

similar way as the consumer’s substitutability with discriminatory retail prices.

Proposition (4.3) shows the prisoner’s dilemma at both wholesale and retail levels: the SPNE

has unilateral wholesale prices and sunk costs of investments without improving payoffs. In

other words, the status quo profit to each network, given by the monopoly retail revenue, is

reduced by the investment fixed cost. Next we explore networks’ temptations to take

advantage of the steering technology.

4.3.3 Roaming Alliance Formation Game

Proposition (4.3) predicts that all networks invest in steering and choose the baseline

unilateral wholesale price. In this case, steering does not effectively take place, because

wholesale prices are symmetric (at ). This section explores incentives to a home network

in engaging in steering unilaterally, or mutually with an IOT partner (i.e. forming a roaming

alliance). For simplicity, the payoffs shown below regard a network’s net payments with its

IOT partners, as we ignore the retail revenues and the fixed cost of steering investment,

which are common to all networks.

Denote for the net payment from an IOT agreement. The subscript refers to the network

in question and the superscript to its partner, with subscript/superscript ( ) for the case

steering is made by a network to the partner, or ( ) for steering away from that partner.

Before we get to the main findings of the game, we state the following Lemmas with their

proofs for the different scenarios of roaming alliance formations.

Lemma (4.5). Assume all networks invested in steering and charged the unilateral wholesale

price. If , where ( ), and if only one network unilaterally steers, it will not

improve its payoffs.

91

Proof. Assume only steers unilaterally to . ’s payoffs are

( )

( )

and

( )

( )

The net gain to is zero.


price. If , where ( ), and if only two networks engage in mutual steering,

they will improve their payoffs.

Proof. Assume only and engage in mutual steering (i.e. form a roaming alliance). ’s

payoffs, for example, are

( )

( )

and

( )

( )

The net gain to is positive.


price. If , where ( ), and if two pairs of networks engage in mutual steering,

the payoffs will be zero to each network.

Proof. Assume and form a roaming alliance and and form another roaming

alliance. ’s payoffs, for example, are

( )

( )

and

( )

( )

The net gain to is zero.


price. If , where ( ), and if one network unilaterally deviates from a mutual

steering, it will not improve its payoffs.

92

Proof. Assume and form a roaming alliance and and form another roaming

alliance, similar to the case in Lemma (4.7). First, assume , for example, deviates by

steering away from . ’s payoffs are

( )

( )

( )

and

( )

( )

( )

The net gain to is still zero. Second, assume deviates by the refrain from steering.

’s payoffs are

( )

( )

and

( )

( )

The net gain to is also zero. Similarly, if only one roaming alliance exists, say between

and , the deviant will not improve its payoffs.

All results of the Lemmas (4.5) to (4.8) require , which is the case where steering can

be effective because visited networks have overlapping coverages. From Lemma (4.5), one

can see that if a network’s unilaterally manipulate visited networks’ market shares, it affects

the payoff with each, but does not improve its total gains. From Lemma (4.6), a mutual

steering improves the payoffs to each member of the roaming alliance at the expense of the

non-members which did not form a counter alliance. From Lemma (4.7), with two roaming

alliances, all gains/losses from net payments cancel out, so the payoffs are zero to each

network. From Lemma (4.8), there is no incentive to a network to deviate from a mutual

steering with its roaming alliance member, whether deviation is by steering away from its

member or by the refrain from steering. Of course, such deviation harms the complying

member.

We conclude from Lemmas (4.5) and (4.8) that only mutual steering can be profitable, which

can sustain roaming alliance cooperation. Members will break even in their own net

payment, where each expects a positive net payment with the non-member due to market

share asymmetry (buying less through steering away compared to selling more at default

market share). Lemmas (4.5) to (4.8) lead to the following proposition.

93

Proposition (4.4). Assume all networks invested in steering and charged the unilateral

wholesale price. If , where ( ), then joining a roaming alliance is a dominant

strategy to each network.

Proof. From Lemmas (4.5) and (4.8), mutual steering is sustained within the roaming alliance

members. From Lemmas (4.6), any network that does not join an alliance will pay a net

payment to its IOT partner that joined an alliance. From Lemmas (4.7), joining an alliance

provides non-negative payoffs to each network.


(4.4). Let ( ) be a possible roaming alliance and ( ) be a possible counter

alliance. Consider the decision in joining different alliances by and . Let represents

their non-zero net payment from their IOT agreements. Table (4.4) shows the payoffs matrix.

7Table (4.4) Payoffs matrix for joining a roaming alliance.

Possible roaming

alliances:

( ) ( )


Join ( )

Don’t join

In Proposition (4.4), the substitutability index, is assumed positive; hence .

This game has zero-sum payoffs, where joining a roaming alliance is its unique equilibrium.

The non-zero net payment if one network joins while the other does not is an externality that

makes joining an alliance a dominant strategy. If only one pair forms an alliance, its

members can improve their statues quo payoffs at the expense of the counter pair.

Therefore, two alliances are formed in equilibrium, and their net gains are zero.

4.3.4 Concluding Remarks

The roaming alliance model in this chapter is based on assuming all networks apply the

same uniform retail price, which is assumed to be the monopoly price. If steering is absent,

the wholesale price also equals the monopoly level. Through investment in steering, a home

network can cause downward pressures on wholesale prices, seeking a gain from the net

payment due to asymmetry in wholesale prices if its IOT-agreement partner does not invest

94

in steering. The equilibrium of this two-stage game is such that each network invests in

steering and wholesale prices are set non-cooperatively (see Table 4.5 below).

8Table (4.5) Standardised prices when all networks apply the monopoly retail price in the steering investment game.

Home network decision to invest in

steering Wholesale cost Retail price

Do not invest in steering

Invest in steering

.

, ).

Therefore, the game does not predict unilateral menu prices as in Lupi and Manenti (2009)

who used a uniform retail pricing different from our model, nor it predicts above marginal

cost pricing as in Bühler (2010) who looked at the impact of wholesale prices on retail two-

part tariffs set by home networks competing for home subscribers.

In the game of joining a roaming alliance, we found only mutual steering is profitable, and

alliance networks can sustain cooperation. Thus, steering is necessary for alliance

formation, as predicted in Bühler (2010) and Hualong and Shoulian (2009). The model

shows that the higher is the substitutability parameter, the higher will be the expected gains

from steering, reflecting the role of coverage overlapping in steering effectiveness. Each

network has a dominant strategy to join an alliance in order to gain from the net payment, or

to balance its net payment with the outsider which joined a counter alliance. The model

predicts two alliances are formed in equilibrium, but do not lead to exclusive IOT agreements

because the substitutability parameter is bounded. With two roaming alliances formed in

equilibrium, the monopoly retail revenues are reduced by steering investment. Such

investment is then a waste of resources as predicted by Lupi and Manenti (2009), and by

Ambjørnsen, et al (2011).

4.4 Discussion

Cross-ownership and alliance models are inspired by the existing literature that addresses

the impact of IOT preferential agreements on game equilibria. In this chapter, our results are

all based on the implications of net payments. In the situation whereby only one pair of

networks has a preferential agreement, the net payment of an IOT agreement between an

95

insider and outsider is non-zero because of asymmetric demands; and, in the cross-

ownership case, because of asymmetric wholesale prices.

In the cross-ownership model, the potential gain from the net payment with the outsider

provides the incentive to join a cross-ownership pair if discriminatory retail policy is assumed

and networks are imperfect substitutes. But the opposite is true with the uniform retail policy

without steering investment: the potential gain from the net payment exists for the outsider

when it does not join a cross-ownership pair. Therefore, with discriminatory policy, vertical

cooperation through cross-ownership is a dominant strategy to each network; while with

uniform policy, non-cooperation is the dominant strategy.

In the alliance model, monopoly retail price is assumed. Each network has a dominant

strategy to invest in steering to induce wholesalers to set unilateral wholesale prices. Each

network expects a gain from the net payment with its IOT partner through wholesale price

differential had that partner did not invest in steering.

With all networks acquiring steering technology and charging wholesale prices non-

cooperatively, a unilateral steering by a home network does not improve its payoffs. Unless

engaged in mutual steering (i.e. roaming alliance), the home network does not improve its

payoffs. Each network has a dominant strategy to join an alliance in order to gain from the

net payment with the non-member IOT-agreement partner had this partner did not join a

counter alliance. Such gain requires visited networks to be substitutable, and rises with

substitutability reflecting the effectiveness of steering.

In this chapter, we found a dominant strategy in each game (cross-ownership and alliance),

which are all driven by the implications of net payments. In all models, networks are

assumed symmetric; thus the choice on a preferred partner does not matter.

Yet since net payment plays a role in each model, a relatively larger network (e.g. with

higher market size) will always be a net payee in its IOT agreements. Therefore, under the

preferential agreements where vertical cooperation is a dominant strategy (i.e. the cross-

ownership with discriminatory retail prices and the roaming alliance), each network will prefer

to choose the largest network as a partner. At the same time, a large network is better off

choosing, as a preferential partner, the largest network amongst its IOT-agreement partners,

in order to minimize negative net payments. We expect the choices of preferential partners

in this case would be: large networks against small networks. Our expectation contradicts

96

with the scenarios analysed in Shortall (2010) and Lacasa (2011), who found small networks

would prefer to form an alliance.

Next chapters will use a dataset on outbound roaming by Etisalat. The dataset contains

wholesale prices and market shares of visited networks. It also contains information on the

visited networks that are cross-owned by Etisalat, as well as the networks that are roaming

alliances to Etisalat. The dataset will be used to test empirically the predictions of the

theoretical models in Chapters (3) and (4) about the base model and its extension.

97

Chapter (5). Dataset Description

In this chapter, we describe the dataset on Etisalat62 (the incumbent mobile network operator

in the UAE) outbound roaming that shall be used for empirical estimations. The purpose is to

introduce the reader to the specifics of the dataset. The dataset contains information which

is relevant in understanding upstream competition, complementary to the theoretical

predictions of previous chapters.

Section (5.1) discusses the applicability of the dataset to the theoretical models of previous

chapters. Section (5.2) provides a brief on Etisalat retail prices. Section (5.3) details the

available information on visited networks. Data collection procedures and dataset description

are provided in Sections (5.4) and (5.5) respectively. We conclude in Section (5.6) with

linking the dataset to the empirical models of next chapters.

5.1 Relevance Of Dataset

The theoretical models in Chapters (3) and (4) focus on wholesalers’ competition in hosting

the visiting subscriber of their Inter-Operator Tariffs (IOT) agreement partner. The home

network is assumed to be a monopolist at the retail level. The games assumed retail price is

either discriminatory (i.e. differs by visited network in the same visited country) or uniform.

The substitutability between visited networks is interpreted as the visited networks coverage

overlap, whereby roamers can choose between visited networks through manual-network-

selection feature of their handsets. Alternatively, home networks can steer to their preferred

networks had they invested in steering technology and set the uniform retail price such that

roamers put their handsets on automatic mode. Preferred IOT agreements are assumed

under cross-ownership or roaming alliance strategies.

The dataset in hand allows us to investigate some of the theoretical predictions of the

previous chapters. It contains all foreign networks that were visited by Etisalat’s travelling

subscribers (hereafter roamers) around the world during the years 2008-2011. It includes

average wholesale prices for the relevant roaming traffics (hereafter quantities), where

wholesale prices are unregulated and hence should represent the market outcome.

62 Etisalat was ranked 140

th among the Financial Times Top 500 Corporations in the world in terms of market capitalization,

expanding in 18 countries (Gulfnews.com, 2009).

98

In addition, visited networks did not price discriminate at the wholesale level between

Etisalat63 roamers (e.g. postpaid versus prepaid), which is expected given a visited network’s

knowledge is supposed to be limited to knowing the home network of the visitors. Therefore,

visited networks are thought to treat an Etisalat roamer as a representative consumer of all

Etisalat roamers.

The study period witnessed Etisalat’s retail policy shift from generally discriminatory retail

prices during the years 2008-2009 to uniform during the years 2010-2011. This policy shift is

supposed to allow Etisalat to steer its roamers to its preferred visited networks as long as

roamers are indifferent between visited networks. We gathered information on visited

networks’ relationships with Etisalat in order to explore the impact of such relationships with

Etisalat on market outcome.

In the UAE, there has been hardly any room for new subscribers when Du, the second

mobile operator, launched its services in 2007, as the market was saturated with mobile

penetration exceeding 100% (Dam, R. 2011). Since there was no mobile number portability

in the UAE during the study period, an Etisalat subscriber would not retain his mobile

number had he switched to Du. If international roaming is mainly about the convenience of

using the same mobile number abroad, we can assume Etisalat subscribers do not switch to

Du for roaming purposes. Therefore, based on the above, we can carry on the theoretical

assumption that the retailer (Etisalat) holds a monopoly position downstream.

5.2 Brief On Etisalat Retail Prices

The NRA of the UAE issued a directive obliging its licenced operators (Etisalat and Du) to

send free SMS to their travelling subscribers informing them about the applicable retail

prices for making and receiving calls. This directive entered into force by the end of 2007

(NRA-UAE, 2007). Therefore, in our dataset, Etisalat’s subscribers are assumed to be aware

of applicable retail prices.

International roaming had been exclusive to postpaid subscribers, but then Etisalat extended

it to prepaid subscribers as the Customized Applications for Mobile Enhanced Logic

(CAMEL) technology has been implemented by Etisalat and many of its IOT-agreement

partners. At the retail level, Etisalat discriminates between its postpaid and prepaid

63 This statement was communicated verbally by Etisalat.

99

subscribers: Etisalat generally charges higher prices to prepaid subscribers compared to the

postpaid. Retail prices vary across roaming services: outgoing and incoming calls, outgoing

SMS and data roaming, whilst incoming SMS is free.

Mobile retail prices are subject to regulation by the NRA of the UAE. Etisalat would submit a

price control request to the NRA, an ex-ante price regulation procedure, only if Etisalat plans

to introduce or renovate prices and packages. However, a retail price change due to

changes in wholesale prices would not be notified to the NRA, including for example retail

international roaming.

Recently, a regional regulation agreement was reached between the GCC member states,

known as the Intra-GCC Roaming Charges. The regulation imposes retail and wholesale

price caps for outgoing calls when roaming within the GCC. The NRA at the UAE side issued

a directive that set a cap on both wholesale and retail prices for outgoing calls regardless of

contract type (postpaid or prepaid) (NRA-UAE, 2010). However, we will show later in this

chapter that wholesale price caps were ineffective during our study period.

From the beginning of the last decade till February 2010, Etisalat offered international

roaming based on discriminatory retail prices, i.e. differentiated by visited networks. The

available discriminatory retail prices are summarized in Table (5.1). For comparison

purposes, the regions are divided into three: the neighbouring GCC countries, the remainder

of the Arab countries and the rest of the world.

9Table (5.1) Etisalat international roaming average retail prices in AED for postpaid roamers in 2009.

Region Incoming

call

Outgoing

local call

Outgoing

call to UAE SMS*

GCC 2.55 (0.35) 1.80 (1.20) 3.43 (1.26) 0.98 (0.43)

Arab countries 3.46 (1.24) 2.54 (1.31) 7.35 (3.35) 1.26 (0.56)

Rest of the world 4.62 (2.36) 2.78 (1.65) 10.26 (4.33) 1.13 (0.57)

All 4.49 (2.31) 2.74 (1.63) 9.89 (4.40) 1.13 (0.57)

*The prices of SMS made to local and international destinations are averaged out for each visited network, which are very similar with most visited networks. The standard deviation in parentheses is then taken for each region. - Standard deviations in parentheses. Source: Adapted from Etisalat (2009).

The standard deviations in the above table give us a hint on how Etisalat set its retail prices

across visited networks within a region. It is obvious that retail prices vary substantially for

outgoing calls (to local or to UAE). This is generally true because, within the region, prices of

outgoing calls vary by visited networks in a given country. With respect to incoming calls,

100

there is variation between regions, but retail price is uniform within most countries, which is

also true with SMS.

Since February 2010, Etisalat applied uniform retail prices by regions: GCC, Arab countries

and the rest of the world (Gulfnews.com, 2010). Table (5.2) shows the (uniform) retail prices

in 201064 for Etisalat postpaid65. As can be seen, prices of incoming and outgoing local calls

tend to be similar, which are about half the prices of outgoing calls to UAE.

10Table (5.2) Etisalat international roaming retail prices in AED for postpaid roamers in 2010.

Region Incoming

call

Outgoing

local call

Outgoing

call to UAE SMS

GCC 1.25 1.30 3.00 1.00

Arab countries 3.00 3.00 6.00 1.50

Rest of the world 4.00 5.00 10.00 2.00

Prices for international outgoing calls to non-GCC destinations are variable. - Any international outgoing call has a setup charge of 2.5 AED. - Optional roaming packages are available for incoming calls and data roaming. Source: Adapted from Etisalat (2010b).

We face difficulties combining information on retail prices with our dataset. First, the

information on retail prices is incomplete. For example, the retail prices in 2008 are not

available; and data roaming retail prices are not available in the years 2008-2009. Second,

retail prices are complex, as they may involve call setup charges and optional roaming

packages (such as for incoming calls and data roaming). Third, for outgoing calls, the

available retail prices differ by call destination (e.g. to local or to international; international to

UAE or to third country), which makes it impossible to combine with our dataset that does

not breakdown the wholesale prices for outgoing calls by destination. Finally, the dataset

would not have enough variations for empirical estimations if we are to use the uniform

(zonal) retail prices in the period 2010-2011.

To overcome the above-mentioned problems, one should use average retail prices (or

revenue per unit). This was requested, but was not supplied as explained next.

64 If Table (5.2) has standard deviations similar to the case in Table (5.1), they would be zero because of uniform retail pricing. 65 Prepaid prices are also uniform (Etisalat, 2010c).

101

5.3 Data Collection

In 2012, we approached the NRA of the UAE for information regarding mobile international

roaming. The NRA shares our interest in knowing whether or not mobile networks compete

in this service, given unjustified high prices that led regional regulators to consider regulation

similar to the EC roaming regulation. In addition, the NRA is interested in knowing the

effectiveness of its existing policies regarding the service, as mentioned in Section (5.2).

We agreed to focus on information regarding UAE outbound roaming; that is, the UAE

subscribers travelling abroad. Such information should be provided at the visited network

level. The choice on years was agreed to be the year after the entry of Du (i.e., 2008), to the

most recent year, which was 2011. Moreover, because international roaming is known to be

seasonal (e.g. summer holiday travels), we agreed to look for annualized data to avoid

seasonality effects. Monetary data was requested to be converted to AED for the relevant

year.

The period 2008-2011 was of relevant importance to the retail pricing policies by Du and

Etisalat: Du began offering uniform retail pricing for some of its services from late 2008

(EITC, 2008); and Etisalat implemented similar retail policy since the beginning of 2010

(Gulfnews.com, 2010).

The NRA issued the following requests for information to its licensed mobile operators,

Etisalat and Du. The requests contain the following items, for each visited network in each

year over the period 2008-2011:

1. Quantities per mobile service (e.g. outgoing calls, incoming calls, etc.);

2. Wholesale total charges in AED;

3. Listed wholesale prices in AED as stated in IOT-agreement;

4. Retail revenues in AED;

5. Listed retail prices in AED as communicated to subscribers (e.g. in website, SMS

notification, etc.); and

6. Number of distinct outbound roamers by subscription type (prepaid and postpaid).

Du’s response was limited to the visited networks in the neighbouring GCC countries for the

period 2010-2011. On the other hand, Etisalat’s response included all visited networks

102

around the world. However, Etisalat did not provide any information on the retail side.

Moreover, neither Etisalat nor Du provided accurate numbers of distinct outbound roamers.

The data provided in Etisalat response66 constitutes our dataset that will be used for the

empirical estimations in the next chapters. Etisalat dataset is comprehensive with regards to

the wholesale level, for it includes all outbound roaming quantities and their respected

average wholesale prices at the visited network level for each of the years 2008-2011. Such

dataset should be relevant for testing the impact of Etisalat’s retail policy shift (from

discriminatory to uniform) on wholesale price competition.

5.4 Information On Foreign Networks

The only information provided in the dataset about foreign networks visited by Etisalat

roamers are the names of those networks, countries in which they operate, and the TAP

code used for billing in international roaming. It is important to note that country information

is not always accurate as some names either had changed, or they did not specify regions

within a country if the country has regional licensing67.

We redefined geographic markets to ensure any derivation on the market-level, such as

market shares, is as accurate as possible. Each visited network is rechecked, through its

website or search engine using its given name and Tap code68, in order to ensure it operates

in the correct geographic market69.

We considered visited networks that have ties with Etisalat (i.e. preferred by Etisalat) for

analytical purposes. We relied on information contained in Etisalat’s response to the NRA

and from Etisalat’s website. In an attachment to its response to the NRA, Etisalat states its

roaming alliance networks (thereafter Alliance networks). By roaming on alliance networks,

Etisalat roamers can buy roaming packages (for incoming calls and data roaming). In

addition, some networks are partially or fully cross-owned by Etisalat (thereafter Cross-

owned networks) (Etisalat, 2011).

66 Disclaimer: Etisalat had an IOT agreement with each visited network in the dataset, which is confidential. Therefore, access to the dataset is restricted to the UAE NRA’s employees. We can show summary statistics and estimation results without disclosing price and quantity information at the visited network level. 67 Regional markets were adjusted for: India (into 23 regions); Iran (into Kish island and Iran); UK (into Great Britain, Isle of Man, Jersey, and Guernsey); Finland (into Alands and Finland); and the disputed region of Karabakh. We lack information to adjust any possible regional markets (e.g. Brazil, Russia and USA). 68 The TAP code is used as a network identifier, which usually differs from the MCC-MNC code (a combination of the Mobile Country Code and the Mobile Network Code) used to hunt information on mobile network operators. 69 See Appendix (G) for full list of visited networks and their geographic markets.

103

Etisalat preferred networks are listed in Table (5.3). Notice that Etisalat cross-owned

networks are in developing countries. Given low mobile penetration, perhaps Etisalat seeks

investment opportunities in markets where potential subscription growth is possible.

Moreover, Etisalat has no more than one cross-owned network in a given market, due

probably to local merger control.

On the other side, most alliance networks are in developed countries, where Etisalat has

more than one alliance in some markets. Given that Etisalat roamers come from the UAE,

one of the world’s highest GDP per capita, perhaps Etisalat seeks efficiency gains in

markets visited frequently by its subscribers.

In addition, we considered visited networks with CAMEL technologies, as they can allow

prepaid roaming. As of 2010, there are 165 visited networks that have CAMEL technologies

and allow roaming by Etisalat prepaid subscribers (thereafter networks With CAMEL)

(Etisalat, 2010a). Table (5.4) lists the countries in which these networks operate.

Furthermore, while gathering information on visited networks, we found that some mobile

operators own more than one network in the same geographic market (e.g. CEDMA and 3G

networks operated by KT in South Korea). This might be a result of network licensing

procedures in some jurisdictions. We also found some visited networks involved in horizontal

merger or acquisition. All such networks are considered as multinetworks within their

markets (thereafter Multinetwork).

An observation in the dataset represents a visited network in a given year, where all visited

networks are segregated by the TAP codes. We do not change how visited networks appear

in the dataset; rather, we control for Multinetwork with the use of a dummy variable70 (one if

a visited network has multinetworks within the market; zero otherwise). Table (5.5) lists

visited networks that have multinetworks.

70 This was suggested by Professor Summit Majumdar, where in Majumdar (2010) and Majumdar et al (2012) a dummy variable equals one if the firm experienced a merger event.

104

11Table (5.3) Etisalat roaming alliance and cross-owned visited networks by country.

Country Alliance network(s)* Cross-owned network**

(Effective year in parenthesis)

Afghanistan - Etisalat (license since 2006)

Armenia Orange -

Austria Orange & T-Mobile -

Bahrain Batelco -

Belgium Mobistar -

Benin - Moov (acquisition since 2005)

Bulgaria Globul -

Burkina Faso - Telecel (acquisition since 2005)

Central African Republic - Moov (acquisition since 2005)

Czech T-Mobile -

Egypt Etisalat Misr Etisalat Misr (license since 2006)

France Orange -

Gabon - Moov (acquisition since 2005)

Germany T-Mobile -

Greece Cosmote -

Hungary T-Mobile -

India Vodafone India Etisalat DB (acquisition since 2008)

Indonesia - Excelcomindo (acquisition since 2007)

Italy Vodafone Italia -

Ivory Coast - Moov (acquisition since 2005)

Jordan Umniah -

Kuwait Zain -

Luxembourg Orange -

Malaysia Maxis -

Moldova Orange -

Netherlands T-Mobile -

Niger - Moov (acquisition since 2005)

Nigeria - Etisalat (license since 2007)

Oman Omantel -

Pakistan - Ufone (acquisition since 2005)

Poland Orange & T-Mobile -

Romania Cosmote & Orange -

Saudi Arabia Mobily Mobily (license since 2004)

Slovakia Orange & T- Mobile -

Spain Orange -

Sri Lanka - Etisalat (acquisition since 2009)

Switzerland Orange -

Tanzania - Zantel (acquisition since 1999)

Thailand AIS & DTAC -

Togo - Moov (acquisition since 2005)

Turkey Turkcell -

UK O2, Orange & T-Mobile -

USA Cingular & T-Mobile -

* Alliance networks were provided as of August 2012. ** Cross-owned networks were provided till the end of 2011 (Etisalat, 2011).

105

12Table (5.4) Number of visited networks with CAMEL technologies by country as of 2010.

Country Networks with

CAMEL Country

Networks with CAMEL

Afghanistan 3 Macedonia 1

Albania 1 Malaysia 2

Algeria 1 Maldives 1

Argentina 1 Mauritania 1

Armenia 2 Mauritius 1

Australia 3 Moldova 1

Austria 1 Morocco 2

Azerbaijan 1 Netherlands 1

Bahrain 2 Niger 1

Bangladesh 4 Nigeria 2

Belgium 3 Oman 2

Benin 1 Pakistan 4

Brazil 3 Palestine 1

Bulgaria 2 Philippines 1

Burkina Faso 1 Portugal 1

Cameroon 1 Qatar 1

Canada 1 Romania 1

Croatia 2 Russia 2

Cyprus 1 Saudi Arabia 3

Czech Republic 3 Serbia 1

Egypt 3 Singapore 1

Estonia 1 Slovenia 1

France 2 South Africa 1

Germany 3 Spain 3

Greece 2 Sri Lanka 3

Hong Kong 1 Sudan 2

Hungary 2 Switzerland 3

Iceland 1 Syria 2

India* 21 Tajikistan 1

Indonesia 5 Tanzania 3

Iraq 3 Thailand 3

Isle of Man 1 Tunisia 2

Italy 3 Turkey 1

Ivory Cost 1 Uganda 2

Japan 1 UK 3

Jordan 3 Ukraine 1

Kazakhstan 2 USA 1

Kuwait 3 Uzbekistan 1

Kyrgyzstan 1 Yemen 2

Libya 1 Zambia 1

Luxembourg 1 Total 165

Source: Etisalat (2010a). * Indian networks with CAMEL operate in 11 regional markets.

106

13Table (5.5) Visited networks with Multinetwork by country.

Country Networks which have Multinetwork

Austria 3G & Vodafone merged in 2009

Brazil Brasil Telco & TNL merged in 2009

Hong Kong

CSL & NW merged in 2006

H3 operates two distinct networks: 2G & 3G

PCCW operates two distinct networks: 2G & 3G

India

In Karnataka market: Spice merged with Idea in 2008

In Punjab market: Spice merged with Idea in 2008

In Rajasthan market: Airtel acquired Hexacom in 2004

In Chennai market: Airtel acquired Skycell in 2000

Iraq Zain acquired Iraqna in 2008

Netherlands Orange & T-Mobile merged in 2007

South Korea KT operates two distinct networks: CDMA & 3G

Taiwan Far East acquired KG in 2003

Tajikistan Tcell operates two distinct networks: North & South; which had operational merger in 2011

UK Orange & T Mobile merged in 2009

Ukraine URS acquired Golden Telecom in 2008, forming Beeline

Beeline & Kyivstar merged in 2010 Source: Information from multiple search engines.

5.5 The Dataset

In our aggregated dataset, we observe foreign networks that were visited by Etisalat

roamers during the years 2008-2011: Etisalat is the home network; and visited networks are

the wholesalers. Visited networks billed Etisalat for the use of their networks. The dataset

includes average wholesale prices (computed by dividing the billed charges by the used

units) for each of the four roaming services: minutes of outgoing calls (to any destination),

minutes of incoming calls, originated SMS (to any destination) and MB of data roaming.

Wholesale prices are available in AED.

These IOT bills are assumed to include any payment discounts given to Etisalat 71 .

Nevertheless, these bills are not the net IOT payments between Etisalat and visited

networks; rather, they are part of Etisalat net payments. More specifically, they represent

Etisalat wholesale costs, or visited networks’ wholesale revenues.

There are few observed visited networks72 that offer mobile telephony via satellite or other

enabling technologies on airplanes or ships, which charged excessive wholesale prices. As

71 Value added tax (VAT) is also included in these bills if imposed by the visited country. However, such tax does not affect wholesale competition because it would be applied similarly to all networks in a given country. 72 Namely, AeroMobile, Maritime Communications, Maritime Seanet, Oceancel, OnAir, Telecom Italia Maritime, Thuraya and Digicel Maritime.

107

they operate in wider geographic markets that include more than one jurisdiction, networks

of this type were removed from the dataset in order to focus on conventional mobile

networks.

We make use of the information provided in Tables (5.3 to 5.5). This information indicates

the relationship between Etisalat and the visited networks in terms of possible preferential

IOT agreements that lead to preferential wholesale pricing and/or steering arrangements. In

addition, we control for multinetworks, because such visited networks may enjoy market

power in their wholesale price setting. Therefore, we created the following dummy variables

to be used for empirical estimations:

Cross-owned by Etisalat: a dummy variable that equals one if the visited network is

cross-owned by Etisalat, zero otherwise. This variable can vary over time (see Table

5.3).

Alliance to Etisalat: a dummy variable that equals one if the visited network is recognized

by Etisalat as an alliance network, zero otherwise. We have this information for 2012

(see Table 5.3), but lack information for the years 2008-201173

. We consider such

networks to be alliances during the full study period; hence the variable is time invariant.

With CAMEL: a dummy variable that equals one if the visited network allows Etisalat

prepaid roamers, zero otherwise (see Table 5.4). We have information for 2010, but lack

information for other years74

. We consider such networks to be with CAMEL during the

full study period; hence the variable is time invariant.

Multinetwork: a dummy variable that equals one if the visited network experienced a

horizontal merger or acquisition, or owned more than one network in the same market,

zero otherwise (see Table 5.5)75

. The dummy variable takes the value of one for each

visited network that has multinetworks. For the merger/acquisition events, the year in

effect is considered; hence the variable varies by year. For example, in the merger event

in the UK between Orange and T-Mobile in 2009, the dummy value is zero in 2008 and

73 In fact, the number of Etisalat alliance networks rose from 38 networks in 2012 as in Table (5.3) to 143 as of July 2013 (Etisalat, 2013b). 74 In fact, the number of networks with CAMEL that allow Etisalat prepaid roamers rose from 165 networks in 2010 as in Table (5.4) to 280 as of May 2013 (Etisalat, 2013a). 75 In Indian regional markets, we noted four merger/acquisition events in Table (5.5). However, our dataset shows the event partners as one network, contrary to other countries where the dataset would still show the event partners as distinct networks. We conclude that those merger/acquisition events took place across-regions, not within region, perhaps to allow one network to enter a regional market. Because such events do not affect the number of existing networks or horizontal competition, none of the Indian networks is considered to have multinetworks.

108

one in 2009-2011 for each network. For the ownership of more than one network (due to

probably licensing reasons) the dummy variable takes the value of one in the full study

period, zero otherwise. For example, in Hong Kong, H3 owns 2G and 3G networks, so

the dummy value is one for the years 2008-2011 for each network.

Summary statistics are provided in Table (5.6) for the dataset variables. We have 552 visited

networks observed over the years 2008-2011; thus the dataset has 2,208 observations. The

number of (active) visited networks that sold any roaming unit increases from 438 in 2008 to

539 in 2011.

Any difference between active networks and the number of observed visited networks, 552,

should mean some networks did not host Etisalat subscribers in the given year. This can be

due to not being licensed that year, or due to other reasons such as Etisalat roamers chose

not to roam on them or Etisalat did not have an IOT agreement with them. It is also possible

that these networks were in markets not visited by the Etisalat roamers in the given year.

The increase in the number of active networks over time indicates the expansion of Etisalat

IOT agreements. This is true despite the existence of Etisalat’s preferred networks (Alliance

or Cross-owned). Therefore, the existence of preferred networks did not lead to exclusivity,

which confirms conventional assumptions that mobile networks are not perfect substitutes.

Our dataset identified 204 distinct visited markets in 182 countries, where some countries

(e.g. India) have regional markets. Etisalat roamers visited 191 in 2008, 193 in 2009, 201 in

2010, and 202 in 2011. Notice on average, each market has 2.7 observed visited networks.

This reflects the oligopoly nature of the mobile industry.

The means in Table (5.6) do not change for Alliance or CAMEL because the dummy variable

does not change over time. In contrast, Cross-owned and Multinetwork vary overtime, but as

can be seen, the mean does not vary by much. Overall, on average, cross-owned networks

represent 5.4% of observations, roaming alliance networks represent 8%, networks with

CAMEL represent 29.9%, and 5.3% of networks own multinetworks within their markets.

Table (5.6) also shows a downward trend in average quantity of outgoing and incoming

calls76. With SMS, apart from the spike in 2009, a downward trend is also apparent. On the

other hand, there is an upward trend in data roaming.

76 The total of outgoing/incoming calls also has a downward trend.

109

14Table (5.6) Dataset summary statistics.

Summary statistics 2008 2009 2010 2011 All

Observations 552 552 552 552 2208

No. of active visited networks 438 448 516 539 1941

Mean of Cross-owned networks by Etisalat (1 if Yes; 0 otherwise)

0.053 0.054 0.054 0.054 0.054

Mean of Alliance networks to Etisalat (1 if Yes; 0 otherwise)

0.08 0.08 0.08 0.08 0.08

Mean of networks with CAMEL allowing Etisalat prepaid (1 if Yes; 0 otherwise)

0.299 0.299 0.299 0.299 0.299

Mean of Multinetwork (networks horizontal ownership) (1 if Yes; 0 otherwise)

0.043 0.054 0.056 0.06 0.053

Outgoing calls

Mean of Quantity (1000'minutes) Redacted Redacted Redacted Redacted Redacted

Mean of Price (AED/minute)* Redacted Redacted Redacted Redacted Redacted

Incoming calls

Mean of Quantity (1000'minutes) Redacted Redacted Redacted Redacted Redacted

Mean of Price (AED/minute)* Redacted Redacted Redacted Redacted Redacted

SMS

Mean of Quantity (1000'SMS) Redacted Redacted Redacted Redacted Redacted

Mean of Price (AED/SMS)* Redacted Redacted Redacted Redacted Redacted

Data roaming

Mean of Quantity (1000'MB) Redacted Redacted Redacted Redacted Redacted

Mean of Price (AED/MB)* Redacted Redacted Redacted Redacted Redacted

* Constant prices, using 2007 as the base year (NBS, 2013). - Data redacted as per NRA-UAE’s data confidentiality policy.

Etisalat roamers travelled to many countries, as the following map suggests. However,

visited markets have high variations in their total quantities sold to Etisalat roamers. Based

on outgoing calls per visited market during the study period, the mean is around 786

thousand minutes, while the median is around 126 thousand minutes. In the next chapter,

we will show how heterogeneity of visited markets, (which is assumed to be a sampling

problem), impacts our estimations. The shading areas of the world map in Figure (5.1)

illustrate the highly visited markets.

110

5Figure (5.1) Map for highly visited markets by Etisalat roamers during the years 2008-2011. - Based on market total minutes of outgoing calls above the median. - The map does not show small size countries or regions such as the Indian regions.

We use CPI deflator77 to adjust prices for inflation. There is no substantial change in the

price movement if, instead, current prices were used78.

In Table (5.6), the (average wholesale) prices a visited network charged to Etisalat for a

minute of outgoing call and a MB of data roaming show a downward trend. The year 2010

witnessed an increase in the means for a minute of incoming call and an SMS. Wholesale

prices differ by roaming service due to probable differences in marginal costs. The highest

price is for a MB of data roaming, followed by a minute of outgoing call.

There are a few price outliers. The main outlier is an Etisalat cross-owned visited network

that charged a very high price for outgoing calls and SMS. The other outliners involve zero

pricing, which can be due to pricing policies. Few visited networks did not charge Etisalat

when usage was very low. This practise is a known bill-and-keep arrangement in

interconnection agreements.

77 The price in year t is multiplied by 100/CPIt. 78 The CPI is 100 in 2007 (the base year), where inflation rose afterwards due to rising property prices. However, during our study period (2008-2011), inflation grew by less than 2%, as follows: CPI is 112.25 in 2008, 114.0 in 2009, 115.0 in 2010, and 116.01 in 2011 (NBS, 2013).

111

However, there are many observations with zero pricing for incoming calls despite high

quantities. This might be due to the fact that with incoming calls, foreign networks are

already compensated by international settlement rates because a roaming incoming call

requires international calling: Etisalat pays such settlement rates to route the call back to the

visited network. This explains why Etisalat sets a positive retail price despite a very low (IOT)

wholesale price.

Visited networks in the GCC charged lower wholesale prices for outgoing and incoming calls

than the average in Table (5.6); while they sold more units than the average in each of the

four roaming services. Although the Intra-GCC Roaming Charges regulation is supposed to

reduce wholesale prices charged between networks in the GCC, we show in Figure (5.2)

that such regulation is ineffective during our study period.

In Figure (5.2), clearly the GCC networks charged Etisalat wholesale prices for outgoing

calls above the GCC caps, except for one network79. This finding shows how regulation

across different jurisdictions is difficult to enforce. Therefore, we can carry on the assumption

that wholesale prices in the dataset are free from regulation.

6Figure (5.2) Wholesale price for a minute of outgoing call by GCC visited networks by year. - The horizontal lines represent wholesale price caps set by the Intra-GCC Roaming Charges for outgoing calls

80.

- The years 2010-2011 are considered because they are relevant in the NRA-UAE’s directive (NRA-UAE, 2010). - Data redacted as per NRA-UAE’s data confidentiality policy.

79 The wholesale price of network 10 in Figure (5.2) is still above the local call price cap. Charging below the international call price cap is not a response to regulation; rather, it is in line with prior periods (before regulation): network 10 increased its wholesale price compared to the previous two years before regulation (i.e. 2008-2009). 80 GCC price caps are converted from SDR to AED according to the effective dates in the NRA-UAE directive (NRA-UAE, 2010), using the SDR Valuation History for the exchange rate of the US dollar ($0.27=1 AED) from the International Monetary Fund website (www.imf.org).

0

1

2

3

4

5

6

2010 2011 2010 2011 2010 2011 2010 2011 2010 2011 2010 2011 2010 2011 2010 2011 2010 2011 2010 2011 2010 2011 2010 2011 2010 2011

1 2 3 4 5 6 7 8 9 10 11 12 13

International call cap 2010

International call cap 2011

Local call cap 2010

Local call cap 2011

112

For later use, we will create a dummy variable for Etisalat uniform retail policy (one for the

years 2010-2011; zero otherwise). This variable will be used as an explanatory variable for

estimating the impact of Etisalat retail policy shift (from discriminatory to uniform) on

wholesale pricing of visited networks and steering by Etisalat.

Furthermore, we look at correlations between the four roaming services in terms of prices,

quantities and market shares (computed by dividing a visited network’s quantity by the total

quantities of the relevant geographic market). This is done to know how interdependent the

services are, especially outgoing and incoming calls, following the EC (2006).

In Table (5.7), there are missing observations in wholesale prices due to some networks

being inactive in some of the years. The missing observations in market shares are due to

some geographic markets having a zero total quantity for some roaming services in some of

the years, mainly for MB of data roaming.

15Table (5.7) Correlations between roaming services.

Correlation of quantities (2,208 observations)

Outgoing

calls

Incoming

calls

SMS

Data

roaming

Outgoing calls 1 - - -

Incoming calls 0.986 1 - -

SMS 0.305 0.320 1

Data roaming 0.566 0.526 0.082 1

Correlation of average wholesale prices (1,344 observations)

Outgoing

calls

Incoming

calls

SMS

Data

roaming



SMS 0.310 0.478 1 -

Data roaming 0.284 0.117 0.419 1

Correlation of market shares (1,851 observations)

Outgoing

calls

Incoming

calls

SMS

Data

roaming



SMS 0.972 0.971 1 -

Data roaming 0.834 0.811 0.800 1

As seen in Table (5.7), all correlations are positive; this concurs with our preliminary intuition

that the four services are non-substitutes. Notice the high significant correlation between the

113

quantities of outgoing and incoming calls, while the correlation of their wholesale prices is

very low, reflecting the gap in prices (i.e., in Table (5.6), the average wholesale price is 6.36

AED for a minute of outgoing call compared to 0.88 AED for a minute of incoming call). We

relate the gap in prices to the involvement of another wholesale price agreement, which is

international settlement rate agreement, as follows.

If an Etisalat roamer makes an outgoing call, the visited network bears the cost of call

origination. In addition, the visited network bears the cost of call termination, which is paid as

an access price if the call is destined to another network. If the terminating network is

located in a different country, such as Etisalat, the visited network has to pay an access

price called international settlement rate, which tends to be above the actual cost of

terminating the call. To cover such costs, the visited network sets high IOTs for outgoing

calls.

In comparison, if an Etisalat roamer receives an incoming call, the visited network bears the

cost of call termination on its own network, without paying any access price. Instead, the

visited network receives an international settlement rate from Etisalat 81 . Therefore, the

visited network charges low IOTs for incoming calls. This explains the gap in wholesale

prices between outgoing and incoming calls.

With roaming incoming calls, Etisalat pays the settlement rates, thus it charges a positive

retail price for incoming calls. If Etisalat does not charge for incoming calls, its subscribers

may view incoming and outgoing calls as substitutes; and we would expect a negative

correlation between quantities of incoming and outgoing calls. The EC (2006) believes that

a home network has an incentive to keep a high retail price for incoming calls; otherwise,

consumers would prefer to receive the call instead of originating it.

Furthermore, the low correlations between the prices suggest differences in actual marginal

costs between the four roaming services. Notice the high correlations between market

shares, suggesting visited networks would have similar market shares across the roaming

services they provide to roamers.

81 This is because the call is routed from Etisalat in the UAE to the visited country. The visited network that handles the incoming call either receives the full settlement rate from Etisalat, or a portion of it if a third party is involved (e.g. the call is routed through another network in the same country which has an international gateway).

114

5.6 Discussion

We will be using the dataset described in this chapter for the empirical estimations in the

proceeding chapters, for the sake of understanding visited networks wholesale competition.

Given that most interconnection prices (including international settlement rates with foreign

networks) are widely regulated, our dataset has information on wholesale prices (IOTs)

assumed to be set freely by market forces. Additionally, the dataset is comprehensive as it

includes all foreign networks visited by subscribers of the same home network. Etisalat

witnessed a shift from discriminatory retail policy to uniform. Such a policy shift is assumed

to make roamers shift the network-selection-modes of their mobile handsets from manual to

automatic, by which steering by Etisalat is assumed possible.

Etisalat retail policy shift is assumed to influence the demands for outgoing calls because

outgoing calls experienced price restructuring as compared to the rest of roaming services.

In addition, outgoing calls were not offered in roaming packages; thus the roamer is

assumed to leave his handset on automatic mode, enabling steering by Etisalat.

In contrast, incoming calls and SMS generally experienced uniform retail (within visited

countries) prior to the policy; while information is unavailable on retail prices for data roaming

prior to the policy. Incoming calls and data roaming are offered in roaming packages that can

be purchased only from alliance networks, thus they may involve network manual selection

by roamers. Moreover, at the wholesale level, outgoing calls have positive wholesale prices

compared to incoming calls that widely involve a zero-pricing policy. Furthermore, quantities

and prices for outgoing calls seem to be less affected by the time trend in Table (5.6),

compared to the rest of roaming services.

The next chapters will focus on quantities and wholesale prices of outgoing calls, similar to

the studies done by the NRAs in the EU. We will test how wholesale competition is affected

by the shift in Etisalat’s retail policy and by the fact that Etisalat has preferred visited

networks. More specifically, we will estimate the demand for visited networks by Etisalat

roamers; and test for the impact of Etisalat’s uniform retail policy shift on wholesale prices

and market shares. The last part will investigate whether steering by Etisalat was exercised,

and to what extent it enabled Etisalat to cause downward pressures on the wholesale prices

it paid to visited networks.

115

Chapter (6). Demands For Visited Networks

International roaming services are governed by Inter-Operator Tariff (IOT) agreements,

which are a type of two-way access/interconnection wholesale agreements. Such

agreements entitle the subscribers of a home network to use their mobile phones while

roaming on foreign networks.

For the aim of understanding wholesale competition under IOT agreements, we make use of

the dataset described in Chapter (5) for empirical estimations in this chapter and the next

two chapters. The dataset has usages and wholesale prices (IOTs) of all foreign networks

visited by Etisalat roamers.

The study period witnessed Etisalat shifting from discriminatory retail pricing policy (during

2008-2009), whereby prices would differ by visited networks, to uniform policy (during 2010-

2011). The two different retail policies82 are the main theme of the theoretical modeling in

Chapters (3) and (4); thus this dataset will be used to test the hypotheses derived from the

theoretical predictions.

In this chapter, we estimate the demand by Etisalat roamers for visited networks to

understand their responsiveness in a foreign land to price and non-price characteristics of

visited networks before and after Etisalat uniform retail policy. The results from demand

estimation will be compared to results from policy-impact analyses of policy on market

outcomes in the next two chapters, where we will explore the relationship of wholesale

pricing (IOTs) and market structure, and the effectiveness of steering in the presence of the

policy.

This chapter is planned as follows. Section (6.0) introduces the structural model to be used

in this chapter. Section (6.1) introduces the dataset. Section (6.2) outlines the empirical

methodology. Demand estimation results are discussed in Section (6.3). The chapter

concludes in Section (6.4).

82 Retail prices are discriminatory or uniform across visited networks, not across roamers.

116

6.0 Simple Logit Demand

In this chapter, we use a structural model to estimate Etisalat roamers’ demands for visited

networks, borrowing from discrete choice models of product differentiation. The simple logit

demand model assumes that demand can be described by a discrete choice model where

prices are endogenously determined by price-setting firm(s). The implied mean utility level

from product i is found by inverting the market share equation, which is then used as the

dependant variable acting in similar way as the observed output quantities of homogenous

goods (Berry, 1994). Consumers’ tastes are assumed homogenous, so the marginal utilities

of product characteristics are identical for all consumers.

The following explanation borrows from section (5.4.2) in Belleflamme and Peitz (2010).

Suppose the consumer can choose between n products in a given market and an outside

good, , which gives him zero utility. Market share for product i can be written as

* +

∑ * +

where all consumers have the same mean utility level, . The mean utility takes the form

where is the price of product , is a vector of observed product characteristics, and

contains all unobservable product characteristics. is considered the error term in the

following transformed market share equation,

( ) ( )

(6.0.1)

where is share of the outside good. Eq. (6.0.1) represents the simple logit demand, which

can be estimated using linear regression. The price elasticities are given by

( ) if (i.e. own price elasticity),

otherwise (i.e. cross price elasticity).

(6.0.2)

Two major limitations of the simple logit model lie in its estimation of elasticities. With respect

to the own price elasticity, the functional form implies that the lower-price firm must enjoy a

higher markup (i.e. elasticity is low if price is low, and high if price is high). In addition, the

substitution pattern among products may be unrealistic (i.e. a product can have identical

cross price elasticities).

117

Alternative discrete models of product differentiation, for instance the random coefficients

logit, overcome the limitation on cross price elasticity. Given our lack of information on

visited networks’ nests/groups 83 , brands fixed effects 84 , and on Etisalat roamers

demographics85, we will follow the simple logit demand model.

6.1 The Dataset

The dataset86 has aggregated traffics (hereafter quantities) used by Etisalat roamers while

roaming on foreign networks, which were annualized to remove the seasonality effects in the

data. These quantities are provided on the visited network level per year for each of the four

roaming services (minutes of outgoing calls and incoming calls, SMS and MB of data

roaming), along with their relevant average wholesale prices.

The dataset has all foreign networks visited by Etisalat subscribers (prepaid and postpaid)

around the world during the years 2008-2011. There are 552 visited networks, and thus we

have 2208 observations. The number of active visited networks is less than 552, which

means some visited networks did not sell any roaming unit in a given year, due to entry/exist

or other unknown reasons.

The four roaming services (outgoing calling, incoming calling, SMS and data roaming) are

assumed to be separate services (i.e. non-substitutes to one another)87, where the roamer

can buy each service from any visited network.

During the study period, Etisalat shifted its retail policy from (generally) discriminatory retail

prices (in the years 2008-2009) to a uniform retail policy (by region or zone) (in the years

2010-2011). We assume outgoing calling is exposed to the effects of the uniform retail

policy. The other services, to a great extent, experienced retail price uniformity during the

whole study period88. Furthermore, incoming calls and data roaming were offered in optional

83 Possible nests can be the radio technology used by visited networks in a given market (e.g. CDMA and GSM; or 2G and 3G). However, any technology would typically handle roaming outgoing calls with similar quality. 84 It is difficult to know the brands for the 552 mobile networks visited by Etisalat roamers, mainly because of language barriers to access their information. Moreover, many mobile networks have multiple names/brands; and we found many networks rebranded themselves (e.g. Orange and T-Mobile to Everything Everywhere in UK; Vimpelcom to Beeline in Russia, etc…). 85 Such information may help to account for the heterogeneity of Etisalat roamers in different visited markets (e.g. in market r, roamers distribution by postpaid/prepaid, business/consumer, male/female, etc…), which is unavailable in our case. For this information to be relevant, it must be provided per visited market. 86 See Chapter (5) for a description of the dataset. 87 See Section (1.1) on definition of roaming and Chapter (2) on literature review. 88 See Section (5.2) on Etisalat retail prices.

118

roaming packages that would apply on alliance networks and hence may require manual

network selection, which would override steering by Etisalat. Therefore, for the sake of all

estimations in this chapter and the next two chapters, we only focus on the data related to

outgoing calls.

In our dataset, we observe average wholesale prices89. During the discriminatory policy

period, retail prices are assumed to be a markup over the (observed) wholesale prices; thus

variation in retail prices should reflect variations in wholesale prices. During the uniform

policy period, retail prices became uniform based on three zones (GCC neighbouring

countries, the rest of the Arab states, and the rest of the world).

We do not use the zonal retail prices in demand estimation because they do not have

enough variation for analysis90. Therefore, we will rely on wholesale prices provided in the

dataset because they vary by visited networks. With respect to the choice between visited

networks in the case of uniform retail price, roamers may be steered by Etisalat to preferred

visited networks.

Visited networks are assumed unconstrained by regulation in choosing their wholesale

prices to Etisalat, because, during the study period, no wholesale regulation existed that

involve Etisalat91. Moreover, visited networks did not price discriminate between Etisalat’s

subscribers (e.g. postpaid versus prepaid)92.

Etisalat roamers are assumed to be aware of retail prices, since Etisalat is obliged by NRA in

the UAE to send SMS to its travelling subscribers informing them of the applicable retail

prices in a given visited market93.

Etisalat retail policy shift is assumed to make its roamers indifferent in choosing between

visited networks based on price, allowing Etisalat to steer to its preferred networks. That is, if

roamers are indifferent, their handsets would be on automatic mode which allows for

steering.

89 Wholesale prices were provided to us in averages: total IOT bill divided by total quantity for each roaming service. 90 A better way is to get average retail prices (total retail revenues divided by total quantity), which was not provided to us. See Section (5.3) on the original data requested from Etisalat. 91 Intra-GCC Roaming Charges regulation within the GCC countries, which addresses wholesale prices charged to Etisalat, was ineffective during the study period (see Section 5.5). 92 This statement was communicated verbally by Etisalat. Visited networks charge Etisalat for the total usage of its roamers. On the other side, Etisalat price discriminates at the retail level, where the retail price is usually lower for postpaid subscribers (see Section 5.2). 93 In line with the thesis focus on wholesale competition, the empirical chapters ignore retail competition in the UAE. In light of the non-existence of mobile number portability in the UAE during our study period, we assume retail competition is absent because the main purpose of international roaming is to use the same home number abroad.

119

6.1.1 Visited Networks Characteristics

Additional to the variables in the dataset of wholesale prices and quantities, we included

dummy variables on visited networks characteristics that are available in Etisalat’s website.

These characteristics are assumed to be the key non-price information relevant to roamers’

decisions in choosing between (foreign) visited networks.

The first characteristic is the networks that are cross-owned by Etisalat, which represent

5.4% of visited networks. These networks usually bear the name of Etisalat (e.g. Etisalat-

Egypt and Etisalat-Nigeria), and thus roamers are assumed to be familiar with cross-owned

networks.

In addition, roamers may care about cross-owned networks during the discriminatory retail

price period. This is because, as predicted in the cross-ownership model (Section 4.1.2), a

cross-owned network should charge wholesale price equivalent to its marginal cost, which

should translate into lower retail prices.

All cross-owned networks operate in developing countries. Among which is Etisalat-DB,

which by itself represents half of cross-owned networks. Etisalat DB operated in 15 regional

markets in India, and has very low market share in each Indian regional market. This is a

structural problem for this specific network as it is new in the Indian mobile industry. The rest

of cross-owned networks were acquired recently by Etisalat or launched with a new mobile

licence. Therefore, cross-owned networks should be in the early phase of operation, where

expansion of coverage rollout might be their priority.

The second characteristic is the networks that are roaming alliance to Etisalat, representing

8% of visited networks. These networks operate in different markets that seem to be

frequently visited by Etisalat roamers. Etisalat would include the name of its alliance

networks in the SMS sent to roamers, and thus roamers are assumed to be familiar with

alliance networks.

By roaming on alliance networks, Etisalat roamers can buy roaming packages (at better

retail prices) for incoming calls and data roaming. Nonetheless, these packages do not apply

to outgoing calls, which is the focus of all empirical estimations. Given independence

between roaming services, when Etisalat shifted from discriminatory retail policy to uniform,

roamers should be indifferent regarding the prices of outgoing calls. We proceed by

assuming limited take-rate for roaming packages, and thus as far as outgoing calls are

120

concerned, the impact of such packages on roamers’ choice between visited networks

during the uniform policy should be very limited94.

The third characteristic is the networks that allow Etisalat prepaid roaming, which represent

around 30% of visited networks. Etisalat prepaid roamers can only roam on these networks,

and such networks must have a CAMEL technology95. As the service is costly to the home

network96, Etisalat charges a higher retail for prepaid roamers compared to postpaid; while

visited networks do not price discriminate at the wholesale level97.

Furthermore, two time related variables are used as non-price characteristics. The first is the

year discrete variable that helps to capture the effect of time. Time may affect demand if

technological improvement improves the quality of network (e.g. coverage). The second is a

dummy variable for the existence of Etisalat uniform retail policy. This policy is likely to be

associated with consumer discretion in demand across visited networks: demand-side

substitution before the policy and supply-side substitution via steering after the policy. In

other words, the policy should help us know why a visited network is chosen98.

6.1.2 Variables

The dataset has 2208 observations, consisting of 552 visited networks in each of the four

years (2008-2011) in 204 different markets. The relevant notations and variables used in

demand estimations are listed in Table (6.1) below.

The year variable, YEAR, is used to control for the effect of time. The policy dummy variable,

POLICY (1 if uniform; 0 otherwise), will be used to estimate the impact of Etisalat shift in its

retail policy (from discriminatory to uniform) on outcomes. POLICY will be interacted with

networks characteristic dummy variables to understand the impact of the characteristics on

demand before and after the uniform policy.

94 For example, the roamer can make outgoing call while roaming on network 1 and receive a call while roaming on network 2. His selection is assumed to be manual when price mattered during the discriminatory policy period. Under uniform policy, the only plausible way not to choose automatic mode based on prices would be for opting in roaming packages. Our assumption regarding the independence between roaming services still hold, but once the roamer manually selected a network to buy a roaming package (which does not include outgoing calls), it is implausible to assume he would switch his handset to automatic in order to make outgoing calls. However, we need the independence of roaming services to hold in order to simplify the modelling; so we proceed with assuming limited take-rate for roaming packages. 95 CAMEL technology is a real-time system that requires the communications of both home network and the visited network to deduct instantly from the balance of the prepaid roamer according to his usage (European Parliament, 2006). 96 This is according to a study by European Parliament (2006). 97 This statement was communicated verbally by Etisalat. 98 The choice between visited networks after the policy will be studied in depth in Chapter (8), where the effectiveness of steering is investigated through interacting the policy with the relative price of the visited network to the market arithmetic mean.

121

For demand estimation, we derive the market shares for the inside good (visited networks in

a given market) and the outside good in the market, using the multiplier explained in Section

(6.2.2).

16Table (6.1) Notations and variables related to estimations.

Notation Description

j Subscript for visited networks, where (j=1,…,552).

t Subscript for year, where (t=2008,2009,2010,2011).

r Subscript for geographic market, where (r=1,…,204).

Time variables Description

YEAR Discrete variable for the study years, where (1=2008, 2=2009, 3=2010, 4=2011).

POLICY Dummy variable equals 1 for the years 2010-2011 (when Etisalat implemented its uniform retail policy); 0 otherwise.

Price variables Description

PRICEjt j’s average wholesale price per minute of outgoing call in year t (in constant AED with 2007 as base year).

Visited network’s characteristic variables

Description

CROSSOWNEDjt Dummy variable equals 1 if j is cross-owned by Etisalat in year t; 0 otherwise.

ALLIANCEj Dummy variable equals 1 if j is a roaming alliance to Etisalat; 0 otherwise.

CAMELj Dummy variable equals 1 if j allows Etisalat prepaid roaming; 0 otherwise.

Market shares variables

Description

j’s market share in minutes of outgoing calls in market r and year t.

Outside good market share in year t.

All prices provided in nominal AED are converted to constant prices using 2007 as the base

year (NBS, 2013). All results do not statistically change if we use other base years, or if we

use nominal prices because inflation is very low during the study period99. In addition, three

observations have extreme price values: two observations have zero prices due to the bill-

and-keep arrangement for very low quantity; and one observation for a cross-owned network

by Etisalat with an abnormal price. These three observations will be excluded in all

estimations of this chapter and the next chapters.

99 Inflation during the study period grew annually by less than 2% (see Section 5.5).

122

6.1.3 Sampling Problem

During the years 2008-2011, Etisalat roamers visited 204 different geographic markets (in

182 countries)100. There are high variations between visited markets in the total quantities

roamers consumed. With outgoing calls, the mean per visited market is around 786

thousand minutes while the median is around 126 thousand.

Later in the chapter, we will compare results for observations below or equal the median

(labeled Low Visited Markets) to results above the median (labeled High Visited Markets),

where we find opposite predictions with regards to roamers’ price sensitivity.

Roamers seem to be price insensitive in the low visited markets, which may suggest that

those markets are hosting a different type of roamers, for example diplomats or

businessmen. Therefore, we are concerned that our dataset includes heterogeneous

observations arising from having two different samples.

To include all observations in a demand estimation, one should consider the use of a good

exogenous101 variable to weight the dataset. Tourism statistics in those visited markets may

not necessarily tell us about the travelling behaviors of the UAE people. According to our

dataset for example, the Afghani market sold more roaming quantities than the markets of

the Channel Islands. In addition, judging from distance to the UAE can be misleading. For

example, the UK market sold more roaming quantities than the Iranian market. By looking at

visited markets, it is not obvious if a macro-level statistics (e.g. GDP per capita) can explain

why markets differ in hosting Etisalat roamers.

The number of UAE travelers per visited markets seems to be a good weight if we are to

include all observations to estimate demand, but is currently unavailable. In the demand

estimation, we will be focusing on highly visited markets.

6.2 Empirical Methodology

This section outlines the empirical methodology used in demand estimation. First, market

definition is explained. Second, the estimation model and hypotheses to be tested are

100 Some countries (e.g. India) have regional markets. 101 Thanks to Professor Eugenio Miravete for this point.

123

presented. Finally, the endogeneity problem is addressed, where instrumental variables are

proposed.

6.2.1 Market Definition

This section defines the relevant market in order to derive market-related variables (e.g.

market size and market shares). We rely on the EC’s definition of this market, which is

“wholesale national market for international roaming on public mobile networks”102 (O.J.,

2003).

In the dataset, we recognize 204 geographic markets103 visited by Etisalat roamers over the

span of the study period. The public mobile networks are the licensed networks in a given

market that provide wholesale roaming services to home networks.

The source of roaming utility for an Etisalat subscriber is assumed to come from consuming

minutes of outgoing calls while roaming abroad with the same home phone number. His

outside options include, for example, calling from public payphones/hotel, or calling with

visitor-SIM cards. Outside options also include the refrain from making roaming outgoing

calls. All such outside options are included in the outside good which gives the roamer zero

utility.

A precondition to making an outgoing call is to first roam on a visited network. Because the

unit of observation in our dataset is the visited network, the product market for Etisalat

roamers is the visited network that sold outgoing call minutes. Etisalat roamers can

substitute between visited networks in the relevant geographic market based on price and

non-price characteristics.

Therefore, for Etisalat roamers, the geographic market is the visited market, and the visited

networks are the product market. Notice our market definition involves Etisalat as the home

network. This is to say that for the same visited networks and market, the demands by

Etisalat roamers is different from the demands by another home network, where the

difference can be, for example, in market sizes and incomes of the roamers. Visited network

j is assumed to set different wholesale prices to different home networks, depending on their

different demands and j’s two-way vertical relationships (i.e. IOT agreement) with each home

102 This definition was known as Market (17) in the EC list of recommendation on relevant markets. In 2007, the EC removed Market (17) from this list after it issued roaming regulation (see Section 2.1.2). 103 See Section (5.4) about the method used to define geographic markets. See also Appendix (G) for a list of all visited networks along with their relevant geographic markets.

124

network, which may involve preferential wholesale prices. This point will be revisited in

details when possible instrumental variables are proposed in Section (6.2.4).

6.2.2 Market Size

Because the aim of all estimations is to understand upstream competition, the market, as

defined in Section (6.2.1), is related to the visited networks in each visited market. Therefore,

market size is the sum of the quantities (i.e. all minutes of outgoing calls) in a given market

and year104.

The outside good is included in the market size for demand estimation. We lack information

on Etisalat roamers’ outside options. Therefore, we use a multiplier to calculate market size.

Due to high variations in quantities between visited geographic markets105, market sizes

need to differ according to visited markets.

In order to calculate market size, we use a multiplier, mt,, to multiply the market quantity in a

visited market, qrt,, which gives the market size in that market, Mrt (r indexes market and t

indexes year). Market share for the outside good will be the same across all geographic

markets.

The choice on the multiplier is justified as follows. First, we look at the typical usage inside

the home country, the UAE, which we consider as what roamers would consume had they

stayed inside the UAE (i.e. did not travel). We use aggregate minutes of mobile-to-mobile

calls made by Etisalat subscribers inside the UAE for each of the study years, denoted by

MTMt. The difference between the typical usage inside the UAE, MTMt, and the aggregate

roaming usage, denoted by Qt, is deemed to be the outside good. The outside good should

reflect what roamers consumed of the outside options (e.g. calling from public payphones

abroad) or their refrain from making roaming outgoing calls. mt is thus derived by dividing

MTMt by Qt.

However, we must attribute some of MTMt for international roaming, because not all Etisalat

subscribers can be assumed to have travelled abroad. A proportion vt of MTMt is attributed

to roaming, where vt is the estimated proportion of roamers in Etisalat total subscribers.

104 If, instead, market size is considered as the aggregate quantities of all the 204 markets, then the demand estimation would focus on demands for visited markets (e.g. hot destinations), not for visited networks which is the focus in understanding upstream competition. 105 See the sampling problem in Section (6.1.3).

125

We use the weights of postpaid/prepaid subscribers in roaming usage106: around 85% of

quantity was made by postpaid subscribers. For each year, we weight the number of Etisalat

subscribers by roaming behaviours (e.g. 85% postpaid; 15% prepaid) and divide them by

Etisalat total subscribers to get vt.

Table (6.2) summarizes the calculations related to market size. As can be seen from the

table, around 22% of Etisalat subscribers are estimated to be roamers during the study

period, which is plausible. The outside good market share varies in each year, and its

difference from one represents the inside good market share for any visited market.

17Table (6.2) Calculations related to market size.

Relevant calculations 2008 2009 2010 2011

Estimated proportion of roamers in total subscribers,

0.216 0.224 0.226 0.230

Market size multiplier:

18.50 20.58 19.54 19.40

Outside good market share:

0.946 0.951 0.949 0.948

6.2.3 Estimation Model

Etisalat roamers’ mean utility from roaming on visited networks is regressed on price and

visited networks’ characteristics (cross-owned, alliance and with CAMEL), which are

observed by the roamers and by the econometrician. Other characteristics of visited

networks observed by the roamers but not by the econometrician enter the error term of the

model. The regressors will also include time variables (year and Etisalat uniform policy

dummy variable), and the characteristics interacted with the policy.

The model for the simple logit demand is given by Eq. (6.1), which is estimated by Ordinary

Least Squares (OLS) and Second Stage Least Squares (2SLS). Subscript for year is

dropped as data is pooled, and market subscript is dropped for convenience.

( ) ( )

( )

(6.1)

106 This information does not belong to our dataset. It is provided by NRA-UAE’s for Etisalat roamers in GCC countries.

126

( ) is the natural log of the market share of visited network ; ( ) is the natural log of

the outside good share; and is the error term. All variables are as explained in Table (6.1).

The product refers to the outside good.

The price used in Eq. (6.1) is the wholesale price charged by visited network to Etisalat in

year t. Any retail price markup, due to probably retail price discrimination between

postpaid/prepaid, is unobserved and hence enter the error term.

As shall be see in the results, the price coefficient is only negative and significant in the

highly visited markets, due the sampling problem discussed in Section (6.1.3). Price

endogeneity problem and proposed instrumental variable are discussed in the next section.

We expect visited networks that are preferred by Etisalat (i.e. cross-owned or roaming

alliance) to charge Etisalat lower wholesale prices, and have higher market shares through

demand-side substitution when retail price is discriminatory. Because steering is assumed

relevant only if retail price is uniform, we expect Etisalat after the policy, if its steering is

effective, to raise the market shares of its preferred networks (i.e. supply-side substitution).

We also expect visited networks that allowed Etisalat prepaid roaming (i.e. with CAMEL), to

have higher market share because they host Etisalat prepaid roamers in addition to its

postpaid.

The choice on the market size multiplier affects the constant term in Eq. (6.1). Nonetheless,

the own price elasticity is sensitive to the definition of the market. If the market is defined

more broadly, the own market share decreases, and, as a result, own price elasticity

increases in absolute value (Slade 2009).

After estimating Eq. (6.1), the own price elasticity is given by

( ) ( )

(6.2)

6.2.4 Endogeneity Problem

This section addresses the endogeneity problem known in demand estimations, such as in

Eq. (6.1), and proposes possible instrumental variables (IVs) to overcome this problem.

127

In estimating a demand model using OLS estimator where price is used as a regressor, the

covariance of price and the error term of the model is not zero. This is because quantity and

price are determined simultaneously in the market, such that a shock in demand or supply

will affect the equilibrium quantity and price. In this case, price is not exogenous, which

means OLS provides inconsistent estimates for the model.

Our dataset lacks a possible instrument for price. Market structure variables, such as the

number of visited networks, affect market shares (e.g. the higher the number of visited

networks, the lower should be the own market share). The error term in Eq. (6.1) involves

the unexplained variations in market shares, which can be affected by the number of

networks. Therefore, the number of networks107 can be correlated with the error term; hence

does not suit to instrument for price.

Another reason that the number of networks can be correlated with the error term of demand

comes from our assumptions on the error term. The unobserved network’s characteristics by

the econometrical are assumed to be captured by the error term. Quality of network

coverage is an example of unobserved characteristics, which can be affected by the number

of networks as a response to competition; hence correlated.

Furthermore, regional dummies, such as GCC and Europe, have effects on prices108. They

can be a proxy for cost at the region-level (i.e. group of markets), but not on the market-level

or network-level109. We sought external information for a better cost proxy, as explained

next.

In industrial organization literature, the price of a product in market (or city) can be an

instrument for the price of the same product in market (Hausman et al., 1994)110. This

requires the prices of the products in markets and to be correlated through cost, and the

price in market ( ) must be uncorrelated to the error term of the demand in market ( ).

The justification is as follows. On the supply-side, the product being sold in two different

markets has a common marginal cost such that a cost shock will shift the prices in both

markets111. On the demand-side, the fact that the product is being sold in different markets

107 Chapter (7) explores market structure impact on wholesale pricing. 108 Chapter (7) explores the impact of networks operating in Europe on their wholesale pricing for non-European networks (Etisalat in this case) for the possibility of a spill-over effect caused by the EC roaming regulation. 109 In fact, if we use region dummies as IVs similar to Etisalat zonal retail prices (GCC, rest of Arab and rest of the world), we get very similar results to the 2SLS in model (6f) of this chapter, but with less significance level for price coefficient (10% level when regional dummies are used compared to 1% level in model 6f). 110 Nevo (2001) uses a similar instrument approach. 111 This may require controlling for market specific supply shifters (e.g. local wages).

128

means the demands are independent, such that a demand shock in one market does not

affect the demand in the other market112.

In our case, we deal with visited networks selling roaming outgoing call minutes. The product

is the visited network in market r (the geographic market) in time t, composing the

observations in our dataset. In order to maintain the same product, we should consider an IV

related to the same visited network. In addition, we assume the visited network faces a

common marginal cost in providing one minute of outgoing call.

We propose that a good instrument for j’s wholesale price per minute of outgoing call (or

IOT, our endogenous regressor) is any other price j charges per minute that involves call

origination. On the wholesale level, examples include the wholesale price per minute j

charges to domestic network k for national roaming by k’s subscriber (governed by national

roaming agreement113), or to foreign network g other than Etisalat for international roaming

by g’s subscriber (governed by international roaming agreement). On the retail level,

examples include retail price per minute j charges to its subscriber for any outgoing call

(domestic or international). Another retail-side example is the retail price per minute g

charges (where g is a foreign network other than Etisalat) to its roamer when using j’s

network.

The proposed instrument is justified on two grounds. First, there is common marginal cost to

network j for any minute of an outgoing call. For example, the cost per minute to originate a

call on network j is similar whether it is done by j’s own subscriber (under any standard retail

service), j’s rival subscriber (under national roaming agreement), or by a visitor roaming on

j’s network (under international roaming agreement). Therefore, the proposed IV and the

endogenous IOT would be correlated via the common cost of call origination.

Second, the proposed IV is assumed to be uncorrelated with the demands of Etisalat

roamers. If the IV is for retail prices for international roaming by a home network other than

Etisalat, then roamers of Etisalat and roamers of that IV’s home network are assumed to

differ in their incomes, market sizes, and marginal utilities 114 regarding j’s (the visited

network) price and non-price characteristics. These two justifications need to hold in order to

proceed with the proposed IV.

112 This may require controlling for market specific demand shifters (e.g. demographics). 113 See Section (1.1) for the difference between national and international roaming agreements. 114 In the simple logit demand, roamers are assumed to have identical marginal utilities within the home network.

129

Among the examples for possible IV, we propose the retail prices of international roaming by

another home network than Etisalat, as they are publically available and can be found at the

network level (which is the product in our case).

The NRA in the Sultanate of Oman provided us with the retail prices of one of its licensed

operator, Nawras, as of August 2013 (NRA-Oman, 2013). Oman has another mobile network

operator, Oman-Mobile, on which we also gathered similar information (Oman-Mobile,

2013). We explored the two Omani networks as IVs for the endogenous regressor (i.e. for

wholesale prices charged to Etisalat).

In line with our focus on wholesale prices charged to Etisalat for outgoing calls (which is not

provided by call destination, e.g. local call versus international call), we choose the retail

prices charged by the Omani networks for calling Oman115 while roaming abroad. This is

because calling Oman should involve more cost components (e.g. international settlement

rates) that reflect cost asymmetry between visited networks and markets. For instance, since

Oman and the UAE are neighbouring countries, routing the international call to Oman or to

the UAE may involve similar costs for the visited networks in a given market. Therefore,

visited networks in a given market are assumed to have similar costs compared to networks

in other markets.

The retail prices of the Omani networks, converted to AED currency, are provided at the

network level for postpaid roamers. In each visited market, Nawras charges uniform retail

price, while Oman-Mobile charges discriminatory retail (i.e. differs by visited networks).

Oman-Mobile retail pricing policy is similar to Etisalat’s policy during the years 2008-2009,

while Nawras is similar to Etisalat’s policy during the years 2010-2011, except that Nawras

does not involve in zonal prices. The retail prices of both Omani networks vary across

markets.

The (observed) wholesale price Etisalat pays to visited network j is assumed to be correlated

with the retail price of an Omani network through the underlying costs j faces (e.g. call

origination cost), but the demands (of Etisalat and an Omani network) for j are assumed to

be uncorrelated.

115 The Omani networks provide retail price information by destination of the call (local within the visited market, or international call to Oman). Nawras also provides information for calling to the rest of the world.

130

The wholesale price j sets is assumed to equal its actual marginal cost116 plus its wholesale

markup. This wholesale markup should be affected by wholesale competition or/and demand

of the home network (e.g. roamers income). The markup may also depend on the

relationship of j and the home network117.

A good IV should be correlated with the endogenous regressor through the common cost.

This crucial assumption is exposed to bias caused by the existence of possible preferential

IOT agreements. As shall be seen in Chapter (7), visited networks’ characteristics

demonstrate IOT preferential agreements (e.g. Etisalat alliance networks charge lower

wholesale prices). However, given our data limitation, we cannot fully control for the bias

with the available information on the characteristics. There can be other causes for

preferential agreements such as the net payments from IOT118 and international settlement

rate agreements, which are not available to us.

One possible way to remove the bias of preferential IOT agreements is to consider, as an IV,

retail prices that are uniform within the visited market, because such prices are neutral to

preferential wholesale agreements within the market. An IV including uniform retail prices

should tell us about cost asymmetries across markets, but it tells nothing regarding cost

asymmetries within the market. However, it should overcome the bias caused by possible

IOT preferential agreements.

On the contrary, an IV including discriminatory retail prices reflects the embedded wholesale

prices. Although it reflects wholesale asymmetries within markets, it is not clear if such

asymmetries correspond to actual cost asymmetries in the presence of preferential IOT

agreements. Put differently, it cannot overcome the bias at the wholesale level stemming

from preferential agreements (e.g. alliance discounts).

In our case, we have retail prices of Nawras, which applies uniform retail policy, and Oman-

Mobile, which applies discriminatory policy. We tried using both as IVs, separately and

jointly, for wholesale prices charged to Etisalat (the endogenous regressor). In First Stage

Least Squares (1SLS) regressions, each Omani network has a positive and significant

coefficient at 1% level, reflecting probably common underlying costs. However, we found

116 Marginal cost includes the actual cost of call origination as explained previously. In addition, it includes the actual cost of termination if the call is terminated on j’s network, or the perceived cost of termination (e.g. mobile termination rate or international settlement rate) if the call is terminated on another network. 117 Section (2.1.1) describes how international roaming came into existence. In 1996, the GSM Association issued its Standard International Roaming Agreement (STIRA), which states that wholesale prices are non-discriminatory. This means j should charge same wholesale price to Etisalat or any other home network. Nevertheless, as found in EC (2000), networks involve in wholesale discounts, such as preferential prices between roaming alliance members. 118 Our dataset only includes one side of the IOT net payment, which is Etisalat wholesale costs, or visited networks ’ wholesale revenues.

131

that only Nawras can be used as an IV in the 2SLS119. Therefore, Nawras retail prices will be

used as the IV for the endogenous regressor.

6.3 Results And Discussion

This section presents and discusses the results for estimating Eq. (6.1) and (6.2). First, we

compare between price results for all observations, low and high visited markets. Second,

we look at the effect of time related variables. Third, we discuss price effect and own price

elasticity of demand. Finally, we discuss the results of visited networks characteristics.

Estimation results are presented in Table (6.3), and marginal effects of policy interaction

terms are presented in Table (6.4).

In Table (6.3), OLS estimator is used for all observations (model 6a) and for observations in

low visited markets (model 6b) to demonstrate the sampling problem, using price and time

related variables as regressors. We then use OLS for observations in highly visited markets

with different specifications (models 6c-6e). Finally, we use 2SLS estimator for the highly

visited markets (model 6f), using Nawras retail price as the IV for the (observed) wholesale

prices charged to Etisalat.

One can see the contradiction of the price coefficients under low versus high visited markets

(models 6b-6c). The coefficient is positive and significant 120 at 5% level for low visited

markets, and negative and significant at 5% level for high visited markets. With all

observations, the coefficient is insignificant (model 6a).

Given our data limitations to find a good weight that should solve the sampling problem

described in Section (6.1.3), we proceed in this section with models related to the highly

visited markets, with focus on comparing results under OLS (mainly model 6e) to 2SLS

(model 6f).

Results for the 1SLS used in estimating the 2SLS are deferred to Chapter (7) which includes

wholesale price estimations. All can be said about 1SLS in this section is that Nawras IV has

a positive and significant coefficient at 1% level (with t-value equals 23.46) in explaining the

119 See Appendix (D) for the evaluation of the two IVs. 120 We found the significance of the positive sign in the price coefficient is caused by not controlling for network’ characteristics. In other words, if we include the characteristics in the OLS for low visited markets, the price coefficient is positive but insignificant (see Appendix E).

132

(endogenous) wholesale prices. The positive sign indicates the correlation between the two

variables rooted in the common (actual) marginal costs faced by the visited network.

As shown in Table (6.3), time is found insignificant in all models. Etisalat uniform retail

dummy variable is negative and significant at 10% level with OLS (model 6e) after including

policy interaction terms. The policy is also significant at 5% level in 2SLS (model 6f). The

significance of the policy in the demand is not clear here, but its impact will become clear in

Chapter (8) as it is related to the effectiveness of steering.

The price coefficient increases in significance and magnitude under 2SLS (6f) compared to

the OLS models. In other words, Nawras IV improves the coefficient on price121. The uniform

retail prices of Nawras might resemble roamers’ perception on retail prices, as compared to

the given wholesale prices122.

In model (6f), we have 273 observations with inelastic demand, representing around one

third of the observations. This contradicts with profit maximizing wholesale price decisions123.

Nevertheless, it is an improvement compared to model (6e), where almost all the

observations have inelastic demands. This follows from using the larger price coefficient in

absolute value with 2SLS compared to OLS, which increases the own price elasticity given

by Eq. (6.2).

121 This improvement is in spite of the loss of observations in model (6f) since Nawras prices are not available for each visited network in the dataset. 122 Etisalat roamers are assumed to be aware of retail prices. Prior to 2010, when prices were discriminatory, if some roamers were aware of only one retail price for using a given visited network, they might consider the same price for using the rest of networks in the same market. If so, then these roamers would act as if the retail had been uniform. In this case, Nawras uniform retail prices better resemble roamers’ perception on prices. 123 Comparison between OLS and 2SLS in terms of the number of inelastic demands are used in Table III of Berry, et al. (1995).

133

18Table (6.3) Simple logit demand models.

Dependent variable: ( ) ( )

OLS 2SLS

All Observations

(6a)

Low Visited

Markets

(6b)

High Visited

Markets

(6c)

High Visited

Markets

(6d)

High Visited

Markets

(6e)

High Visited

Markets

(6f)

CONSTANT -4.266*** -3.856*** -4.305*** -5.112*** -5.006*** -3.865***

(0.168) (0.201) (0.253) (0.245) (0.252) (0.332)

YEAR

-0.043 -0.102 -0.028 -0.006 -0.007 -0.060

(0.082) (0.095) (0.125) (0.116) (0.116) (0.129)

POLICY (Dummy=1 for 2010-2011)

-0.349* -0.310 -0.379 -0.328 -0.526* -0.639**

(0.184) (0.212) (0.282) (0.262) (0.290) (0.319)

PRICEjt (Network own price)

0.012 0.038** -0.056** -0.055*** -0.052** -0.203***

(0.014) (0.016) (0.023) (0.021) (0.021) (0.037)

Network characteristics:

CROSSOWNEDjt (Dummy=1 if network is cross-owned by Etisalat)

- - - -0.912*** -0.003 -0.342

- - - (0.285) (0.504) (0.552)

ALLIANCEj (Dummy=1 if network is alliance to Etisalat)

- - - 0.621*** 0.465* 0.439

- - - (0.177) (0.248) (0.274)

CAMELj (Dummy=1 if network allows Etisalat prepaid)

- - - 1.328*** 1.092*** 1.042***

- - - (0.119) (0.174) (0.195)

Interaction with POLICY:

CROSSOWNEDjt

- - - - -1.286** -1.103*

- - - - (0.610) (0.659)

ALLIANCEj

- - - - 0.293 0.133

- - - - (0.353) (0.390)

CAMELj - - - - 0.407* 0.563**

- - - - (0.240) (0.266)

Observations 1,927 932 995 995 995 831

R2 0.015 0.039 0.017 0.158 0.164 -

R2 Adjusted

0.0137 0.0360 0.0141 0.153 0.157 -

Observations with Inelastic Demands

- - - - 993 273

Standard errors in parentheses * p<0.10, ** p<0.05, *** p<0.01

134

With 2SLS (model 6f), the mean of own price elasticity is (-1.185), where the absolute value

of elasticity declines over the four years (2008-2011). This is caused by higher density of

inelastic observations over the years124. As a consequence, the distribution of elasticity is

flatten overtime as shown in Figure (6.1).

The negative trend of own price elasticity can be read as a shift towards inelastic demands,

which is accompanied by wholesale price decline as demonstrated in Chapter (5) and found

in Chapter (7).

7Figure (6.1) Own price elasticity by year. - Based on Eq. (6.2) using the price coefficient in model (6f).

The results regarding visited networks cross-owned by Etisalat imply that cross-owned

networks reduce the mean utility of Etisalat roamers. The coefficient of the dummy variable

is negative and significant at 1% level in model (6d). In models (6e) and (6f), where the

interaction terms enter the model, we found that before the policy, the coefficient is

insignificant, while after the policy, the coefficient becomes negative and significant at 1%

level (see Table 6.4).

124 The number of observations with inelastic demands is 273. By breaking it by year, we have 51 in 2008, 61 in 2009, 72 in 2010, and 86 in 2011.

135

The intuition here is that, only after the policy, Etisalat roamers have less mean utility when

roaming on Etisalat cross-owned networks125.

A possible explanation for the negative utility is that most cross-owned networks are in the

expansion stage since they are either new or have very low market share due to structural

problems (unrelated to roaming). The reduction in mean utility is probably caused by

roaming on networks that have low-coverage (i.e., low quality).

19Table (6.4) Marginal effects on visited networks’ characteristics post Etisalat uniform retail policy.

Network’s Characteristic

Null Hypothesis Relevant

model F-statistics P-value

Cross-owned by Etisalat

OLS (6e) 14.05 0.000***

2SLS (6f) 16.00 0.001***

Alliance to Etisalat OLS (6e) 9.14 0.003***

2SLS (6f) 4.28 0.039**

With CAMEL OLS (6e) 83.02 0.000***

2SLS (6f) 78.68 0.000***

* p<0.10, ** p<0.05, *** p<0.01 - Based on Table (6.3) results.

The alliance dummy variable has a positive and significant coefficient at 1% level in model

(6d). When the interaction terms enter model (6e), the significance level of alliance becomes

10% before the policy and 1% level after the policy (see Table 6.4). On the other side, in the

2SLS (model 6f), alliance is positively related to mean utility, but only significant after the

policy (at 5% significance level as shown in Table 6.4).

The intuition here is that alliance networks increased the mean utility of Etisalat roamers,

mainly after the policy. A possible interpretation of this result is that Etisalat chose roaming

alliance networks that have strong coverage, such that steering towards them (after the

policy) raises roamers’ utilities126.

The findings regarding alliance networks are similar to the ones regarding networks with

CAMEL that allow Etisalat prepaid roaming. However, the magnitude of the effect is much

125 We shall see in the next chapters that Etisalat steers towards its cross-owned networks; despite they do not give Etisalat wholesale discounts. 126 It could be also possible that roamers choose alliance networks to buy roaming packages for incoming calls and data roaming, even if the price for outgoing call is identical for any network in the market. Interdependence between roaming services is assumed away as explained in Section (6.1.1).

136

larger with CAMEL, as can be seen in the size of the coefficient in Table (6.3). The dummy

variable for CAMEL accounts for visited networks that can host Etisalat roamers with prepaid

subscriptions; thus it increases demand through raising the market size.

6.4 Concluding Remarks

In the structural demand model, we focus on highly visited markets because of our concern

that the dataset has sampling problems. Roamers are found price sensitive in highly visited

markets. These markets seem to be visited by many Etisalat roamers of different types. We

found visited networks with CAMEL raise the mean utility of roamers more significantly than

the rest of networks’ characteristics, indicating the presence of prepaid roamers.

On the other hand, the low visited markets have probably certain types of consumers, who

seem to be price insensitive. In fact, when visited networks’ characteristics are included in

the estimation for low visited markets, CAMEL is found insignificant before the uniform

policy, and significant after the policy with positive but small coefficient compared to the

highly visited markets (see Appendix E). This suggests that low visited markets do not

significantly host prepaid roamers. We conclude that roamers in those markets have mainly

postpaid phones, who might be businessmen, journalists or diplomats.

Furthermore, Etisalat preferred networks (alliance and cross-owned) have opposite effects

on mean utility, where they matter only after the policy. Alliance networks raise mean utility;

while cross-owned networks reduce it. As the policy enables steering, Etisalat might be

steering its roamers towards its preferred networks, despite reducing its roamers utility when

they roam on cross-owned networks.

We used Nawras retail prices to instrument for wholesale prices charged to Etisalat. Nawras

applies uniform retail prices that vary across markets, and this variation is assumed to proxy

cost differentials between markets. We think an IV with uniform price overcomes the bias in

wholesale prices caused by preferential wholesale agreements or any other non-cost based

reasons, compared to discriminatory retail price.

We employed a structural model to estimate demand for visited networks. The mean own

price elasticity is found to be (-1.185) in the highly visited markets. For the EEA networks,

137

the figure is (-1.295). Appendix (F) lists the mean own price elasticity of demand for the EEA

markets, where the majority of those markets have elastic demands.

These results regarding Etisalat roamers’ own price elasticity suggest demand for outbound

roaming is more elastic than the industry crude studies by EC (2006), GSMA (2008), Europe

Economics (2008) and CMT (2009) surveyed in Section (2.3.3). However, the focus of those

industry studies is outbound roaming within the EEA networks, which excludes Etisalat.

Next chapters will explore visited networks’ wholesale prices and steering by Etisalat, taken

into account visited networks’ characteristics. We make use of the panel structure of the data

to understand the effect of Etisalat uniform retail price policy.

138

Chapter (7). Visited Networks Wholesale Pricing

In the theoretical chapters (3) and (4), we present pricing games for networks with IOT

agreements, assuming two countries, each with two networks. With discriminatory retail

policy, the unilateral wholesale price is predicted to decline by visited networks’

substitutability level; and between vertically cross-owned networks, wholesale price is set at

marginal cost. If the uniform retail policy chosen at the monopoly level, all networks are

predicted to invest in steering when the substitutability level is high enough to reduce

wholesale prices, which brings efficiency gains. Then networks form roaming alliances in

vertical pairs, where each pair engages in reciprocal steering to raise members’ market

shares.

This chapter tests empirically for the theoretical predictions regarding the IOTs (or wholesale

prices). The chapter studies the effect of market structure, as a measure of competition, on

wholesale prices charged by visited networks to Etisalat. In addition, the chapter studies the

impact of visited networks’ characteristics on pricing, especially how Etisalat is charged by

its preferred visited networks before and after the existence of its uniform retail policy that

would enable its traffic steering.

In this chapter, the dependent variable is wholesale price; hence the model shares similarity

with the 1SLS used in Chapter (6). The model in this chapter does not re-estimate the supply

model (1SLS), because it is concerned with the effect of market structure on prices;

therefore, it will consider all observations (both high and low visited markets).

Furthermore, we are interested to see if visited networks operating in the EEA were

overcharging Etisalat, compared to the rest of networks in the dataset. This is because EEA

networks experienced the EC roaming regulation, and given unregulated wholesale prices

charged to Etisalat, EEA networks might involve in spill-over (or waterbed) effect due to their

regional regulation127.

This chapter is planned as follows. In Section (7.1), the estimation model and hypotheses to

be tested are presented. Then estimation results are discussed in Section (7.2). The chapter


127 Thanks to Professor Tommaso Valletti for this suggestion.

139


The aim of this chapter is to test the theoretical predictions in light of the impact of Etisalat

uniform retail policy, relaying on the same dataset described in Chapter (5) and used in

Chapter (6) for demand estimation. Therefore, we will be using panel regressions. The

random-effect regression will be looking at the effect of time invariant variables; for instance,

to test for the spill-over effect in the EEA. Fixed-effects and random-effects models will be

compared using Hausman test. Due to visited networks entry and exist, the panel is (weakly)

unbalanced.

The relevant variables used in the estimations of this chapter are listed in Table (7.1) below.

We use two variables for market structure. The first variable is NETWORKSrt, which is the

number of active visited networks in market r and year t. This variable is calculated for each

market by summing up all visited networks (in each year) that sold any roaming unit to

Etisalat roamers.

The second variable is MULTINETWORKjrt, which is a dummy variable with value one if visited

network j in market r and year t experienced horizontal merger/acquisition event, or if it owns

more than one network in the same market128; zero otherwise129.

Networks which own more than one network in the same market (or have multinetworks)

represent 5.3% of visited networks in the dataset. The dummy variable controls for local

synergy and thus it is assumed to have the opposite effect to the number of visited networks.

The wholesale price estimation model is given by the following equation.

( )

( ) ( ) ( )

(7)

128 This is probably due to licensing reasons in the market. For example, a network has two licensed networks, one with 2G technology and one with 3G technology. 129 As explained in Chapter (5), the data given to us breaks down observations by the name of visited network (which has unique TAP code) and years. We do not change how the data looks; instead, we control for multinetworks with a dummy variable as suggested by Professor Summit Majumdar.

140

where the ( ) is the natural log of price; and is the error term. All variables are as

listed in Table (7.1).












Description




Market structure variables

Description

NETWORKSrt Number of active visited networks in market r and year t.

MULTINETWORKjrt Dummy variable equals 1 if j owns more than one network in market r and year t; 0 otherwise.

Region variables Description

EEAj Dummy variable equals 1 if j operates in the European Economic Area (EEA); 0 otherwise.

Eight regressors in Eq. (7) appear also in the demand estimation of Eq. (6.1). We will be

reading the results of Eq. (7) in light of the 1SLS results used in demand estimation of Eq.

(6.1). In other words, the price model in Eq. (7) can be compared to the reduced form of the

supply model used as the 1SLS. However, there are two main problems that may lead to un-

similar results.

141

First, the 1SLS pools the data, whereas the price model in Eq. (7) uses panel regression.

Second, the market structure variables, NETWORKSrt and MULTINETWORKjrt, are not assumed

relevant in the 1SLS (see Section 6.2.4); but they are assumed relevant in this chapter for

testing the effect of market structure on wholesale competition. We will present results of the

1SLS next to the random-effect panel regressions that uses same specification130.

The coefficients related to market structure are assumed to be measures of wholesale

competition. As NETWORKSrt (the number of visited networks) increases, wholesale price is

expected to decrease, reflecting the rise in the bargaining position of the buyer (Etisalat).

In contrast, MULTINETWORKjrt dummy variable (one if the network owns more than one

network in the same market; zero otherwise) is expected to do the opposite effect on

wholesale price, because a mobile network operating multiple networks should relatively

enjoy more market power.

Furthermore, we explore whether EEA networks overcharged Etisalat due to possible spill-

over effect of the EC roaming regulation.

In addition, we expect the coefficients related to Etisalat preferred networks (cross-owned or

alliance networks) to be negative. That is, preferred networks are assumed to lower their

wholesale prices to their affiliate, Etisalat.

Moreover, the fact that some visited networks have CAMEL technology (and allow Etisalat

prepaid roaming) is expected to be irrelevant in wholesale pricing decision. That is, visited

networks do not involve in price discrimination at the wholesale level between prepaid and

postpaid roamers.


This section discusses the results from estimating the wholesale price model for Eq. (7).

First, results from the 1SLS regression are discussed, and compared to the random-effect

panel regressions. Then we expand the analysis to focus on wholesale price competition by

including the market structure variables and the EEA dummy variable using random and

fixed effect panel regressions.

130 Fixed-effect panel regression will drop Nawras IV because Nawras IV does not vary by time (i.e. observed in 2013).

142

Table (7.2) presents the 1SLS in model (7a), which uses pooled OLS for observations in

highly visited markets. The table also presents the random-effect panel regressions for

comparison purposes (model 7b for highly visited markets and 7c for all observations). In all

these models, we use the given wholesale price (in level-form) as the dependent variable,

which is also used in the 1SLS. Additionally, we include Nawras IV (as a market-level cost

proxy) 131 and the other explanatory variables used in the 1SLS.

It seems that model (7a) differs because of the choice on the estimator (Pooled OLS versus

random-effect panel regression) or because of the choice on observations based on type of

market (highly visited or all markets). The only discrepancy is in the coefficient related to

cross-owned networks before the policy, which is negative and significant at 10% level. We

found this is caused by the functional form of the dependant variable. That is, if wholesale

price is in log-form instead, the coefficient becomes insignificant132.

More importantly, Nawras IV is positive and significant at 1% level in all models of Table

(7.2). The positive relation confirms the assumption that the two home networks (Etisalat and

Nawras) are correlated through the actual costs borne by visited networks to handle

outgoing calls.

Next we expand model (7c) by including the market structure variables and the EEA dummy

variable to estimate Eq. (7), where the results are presented in Table (7.3). The marginal

effects of policy interaction terms are presented in Table (7.4). Notice the results from model

(7c) are similar in significance to model (7d) of Table (7.3).

131 Explanation for using Nawras retail prices as an IV is provided in Section (6.2.4). 132 The log makes price outliers less significant. Since cross-owned networks are all in developing countries (see Section 5.4), the significance of their price perhaps reflects price outliers.

143

21Table (7.2) Wholesale price models with wholesale price in level-form.

Dependent variable: PRICEjt

High Visited Markets All observations

OLS (1SLS)

(7a)

Random-effect

(7b)

Random-effect

(7c) CONSTANT

2.733*** 2.784*** 2.505***

(0.314) (0.331) (0.262)

YEAR

-0.251* -0.252** -0.284***

(0.148) (0.102) (0.073)


0.655* 0.685*** 0.462***

(0.369) (0.256) (0.177)



-1.050* -0.208 -0.456

(0.635) (0.650) (0.578)

ALLIANCEj (Dummy=1 if network is an alliance to Etisalat)

0.671** 0.713* 0.370

(0.315) (0.389) (0.351)


0.575** 0.471* 0.154

(0.225) (0.267) (0.222)


CROSSOWNEDjt

1.513** 0.584 1.073**

(0.760) (0.574) (0.499)

ALLIANCEj

-1.124** -1.114*** -1.057***

(0.448) (0.307) (0.244)

CAMELj

-0.489 -0.400* -0.212

(0.307) (0.217) (0.160) NAWRAS IV:

NAWRASj (Nawras retail price in AED in 2013 for roaming on j)

0.262*** 0.259*** 0.327***

(0.011) (0.018) (0.015)

Observations 831 831 1,403 Visited Networks 241 238 396 R

2 0.413 - -

R2adjusted 0.406 - -

rho - 0.541 0.611

Standard errors in parentheses * p<0.10, ** p<0.05, *** p<0.01.

In Table (7.3) below, one can see that time has a negative and significant coefficient at 1%

level, which shows the negative trend in real wholesale prices. In model (7g) for instance,

price declines by 5.2% a year.

144

22Table (7.3) Wholesale price models with wholesale price in log-form for all observations.

Dependent variable: Ln(PRICEjt)

Random-effect Fixed-effect

(7d) (7e) (7f) (7g)

CONSTANT

1.410*** 2.031*** 2.069*** 2.252***

(0.074) (0.049) (0.047) (0.047)

YEAR

-0.068*** -0.056*** -0.056*** -0.052***

(0.016) (0.013) (0.013) (0.013)


0.091** 0.078** 0.078** 0.091***

(0.040) (0.032) (0.032) (0.031)



0.135 -0.058 -0.091 0.099

(0.142) (0.119) (0.119) (0.333)

ALLIANCEj (Dummy=1 if network is an alliance to Etisalat)

0.120 0.036 0.096 -

(0.094) (0.097) (0.095) -


0.035 -0.009 -0.013 -

(0.057) (0.056) (0.057) -


CROSSOWNEDjt

0.277** 0.079 0.079 0.028

(0.113) (0.078) (0.078) (0.078)

ALLIANCEj

-0.343*** -0.303*** -0.303*** -0.298***

(0.055) (0.048) (0.048) (0.047)

CAMELj

-0.034 -0.020 -0.020 -0.017

(0.036) (0.029) (0.029) (0.029) Market structure:

NETWORKSrt (Number of networks)

-0.120*** -0.087*** -0.088*** -0.146***

(0.011) (0.009) (0.009) (0.013)

MULTINETWORKjrt (Dummy=1 if network owns more than one network)

0.360*** 0.383*** 0.374*** 0.241**

(0.089) (0.082) (0.082) (0.115)

European Economic Area (EEA): EEAj (Dummy=1 if network operates in EEA)

0.040 0.179*** - -

(0.066) (0.062) - -

NAWRAS IV:


0.059*** - - -

(0.004) - - -

Observations 1,403 1,927 1,927 1,927 Visited Networks 396 549 549 549 rho 0.695 0.778 0.780 0.823

Standard errors in parentheses * p<0.10, ** p<0.05, *** p<0.01.

145

23Table (7.4) Marginal effects on visited networks’ characteristics post Etisalat uniform retail policy.

Network’s Characteristic

Null Hypothesis Relevant model F-statistics P-value


Random-effect (7d) 14.11 0.000***

Random-effect (7e) 0.04 0.838

Random-effect (7f) 0.01 0.915

Fixed-effect (7g) 0.15 0.700

Alliance to Etisalat

Random-effect (7d) 5.58 0.018**

Random-effect (7e) 7.68 0.006***

Random-effect (7f) 4.78 0.029**

With CAMEL

Random-effect (7d) 0.00 0.977

Random-effect (7e) 0.27 0.600

Random-effect (7f) 0.33 0.564

* p<0.10, ** p<0.05, *** p<0.01 - Based on Table (7.3) results.

On the other hand, the uniform dummy variable has a positive and a negative coefficient in

all models, with 5% level in the random-effect models and 1% with the fixed-effect model. In

model (7g) for instance, wholesale price rises by 9.1% after the policy, which undermines the

effectiveness of Etisalat steering to bring visited networks to the negotiation table.

It is clear from Table (7.3) that all visited networks characteristics are insignificant in

explaining the wholesale price before the policy. After the policy, some of the preferred

networks become significant (cross-owned and alliance), while networks with CAMEL are

still insignificant.

After the policy, cross-owned networks are only significant in model (7d), with 1%

significance level (see Table 7.4). Model (7d) suggests that cross-owned networks

overcharged Etisalat. We can see that this result holds only when Nawras IV is included in

the regression. A possible interpretation is provided below.

We think cross-owned networks were in the expansion phase of their network roll-out, so

their priority should be getting financed. Model (7d) suggests that once we control for

market-level costs (with the help of Nawras IV), some cross-owned networks, perhaps with

more need for finance, overcharged Etisalat when Etisalat became in charge of substitution

between visited networks (i.e. via steering after the policy).

146

All results regarding cross-owned networks in Table (7.3) point to one conclusion: cross-

owned networks do not offer wholesale discount to Etisalat. This suggests non-cooperative

pricing behaviour by the cross-owned networks, which is contrary to our theoretical

predictions, despite Etisalat’s ability to steer after the uniform policy133.

On the other hand, after the policy, Etisalat alliance networks lowered their wholesale prices

significantly in the random-effect models (at least with 5% level as in Table 7.4). This is

confirmed by the fixed-effect model, which shows that after the policy, the alliance networks

reduced wholesale prices by 29.8% (at 1% significance level). The institution here is that

alliance networks involved in preferential wholesale pricing with Etisalat when Etisalat is able

to steer.

With respect to networks with CAMEL, we conclude that allowing prepaid roaming does not

matter in the setting of wholesale prices, before or after the policy. This result is in

accordance with the fact that visited networks did not price discriminate at the wholesale

level for hosting Etisalat prepaid/postpaid subscribers.

Table (7.3) suggests that market structure matters in the setting of wholesale prices. The

number of active visited networks decreases the wholesale price at 1% significance level in

all the models. This result reflects wholesale price competition between (substitutable)

visited networks, because Etisalat can increase its bargaining position. In model (7g) for

example, we conclude that the price declines by 14.6% for each extra visited network.

On the other hand, the coefficient on the dummy variable for horizontal multinetworks (one if

visited network owns more than one network within same market; zero otherwise) has a

positive coefficient, which is significant at 1% level with all the random-effect models, and

5% with the fixed-effect model. Owning more than one network in the same market is

assumed to raise market power, which does the opposite effect of the number of visited

networks. In model (7g) for example, we conclude that the price rises by 24.1% for visited

networks with horizontal ownership.

The coefficient on the EEA dummy variable (one if EEA; zero otherwise) is positive and

significant in model (7e) at 1% level134. The positive relation associated with being an EEA

network and wholesale price might reflect spill-over effect.

133 In spite of not getting wholesale discount, Etisalat steers towards its cross-owned networks, as predicted by the steering model in Chapter (8). 134 The sing and significance level of the coefficient do not change at different specifications, starting from the specifications as in model (7e), then gradually removing each regressor except EEA.

147

However, comparing to model (7d), the above result is true only if Nawras IV is excluded

from the model. Nawras IV is a market-level cost proxy, but at the same time Nawras is not

immune from possible EEA spill-over effect, because, similar to Etisalat, Nawras operates

outside the EEA135.

The high wholesale price associated with EEA networks might reflect higher actual marginal

costs in EEA markets, for which we have no cost data to control for. An ideal way to test for

spill-over in EEA networks is to compare wholesale prices before the EC roaming regulation

to after the regulation. This would require data on prices before 2007, the year when

regulation entered into force. Given the study period in our dataset is post EU regulation, we

cannot judge if EEA networks involved in spill-over.

Finally, we compare between the random and fixed effect models (7f and 7g) 136 . The

Hausman test (with the null hypothesis that no difference between the estimated coefficients

of fixed-effect and random-effect models) has a chi2 of 29.5 (significant at 1% level). Hence,

we reject the null hypothesis and conclude the fixed-effect model is preferred over the

random-effect model. Nevertheless, as obvious in Table (7.3), there is no contradiction in the

significance level or sign of the coefficients under both the random and fixed effect models.


This chapter studies the effects of market structure on wholesale prices (IOTs) set by visited

networks to Etisalat. The chapter also explores the effects of visited networks characteristics

on wholesale pricing, and whether EEA networks overcharged Etisalat.

A key finding is that wholesale prices decline by the number of visited networks. An

additional network reduces price by 14.6%, reflecting the increase in Etisalat negotiation

ability. This finding confirms our theoretical models (in Chapters 3 and 5) that networks are

substitutes, where substitutability is our measure of competition that drives down wholesale

135 If, instead, Nawras IV (weather logged or non-logged) is regressed on EEA (with or without the rest of regressors), EEA is found to be have a positive and significant coefficient, at 1% level. 136 These results do not change in significance level if ALLIANCEj and CAMELj are excluded from model (7f) such that models (7f) and (7g) have the same regressors.

148

price. On the contrary, the existing theoretical literature predicts that wholesale price would

rise with the number of visited networks137.

Wholesale competition in IOT agreements is also found in some of the competition analyses

carried out by the European NRAs (i.e., Austria, Italy, and Spain), as summarized in Table

(2.1).

In addition, we found merger/acquisition or any form of horizontal ownership permits

overcharging. Moreover, networks operating in EEA are associated with higher wholesale

price, which can be caused by waterbed effect due to the EC roaming regulation, or due to

having higher actual costs.

Within all visited networks’ characteristics, only alliance networks offer wholesale price

discount, where the discount is offered only after the policy. This result may suggest that

steering by Etisalat is effective, as shall be explored next chapter. Wholesale discount is also

found by the NRAs in Austria and Spain, where the Austrian NRA found significant

wholesale discounts given to alliance networks (see Table 2.1).

Unlike alliance networks, cross-owned networks do not lower their wholesale prices. Such

non-cooperative pricing might be explained by their financial needs for network-expansion.

In addition, CAMEL technologies do not matter in wholesale pricing. This is interpreted as a

reflection of non-price discrimination at the wholesale level that is known in this industry.

137 See Section (2.3.1), where theoretical models in existing literature use demand that makes visited networks appear as perfect complements.

149

Chapter (8). Effectiveness Of Steering

When a home network obtains traffic steering technology, it can affect the market shares of

visited networks. We think three conditions are necessary for steering to be effective. The

first is that the roamer does not obstruct steering, which means his handset must be on

automatic mode. This is assumed to be the case when the retail price is uniform138, and the

roamer cares mainly about the final price he pays. Second, the roamer is in coverage

overlapping areas, where visited networks are substitutable. Third, there are efficiency gains

to justify the home network’s investment decision in the steering technology.

In Section (4.3), we modelled a game whereby all the three conditions are met. When visited

networks are substitutable and home networks set uniform retail price at the monopoly level,

wholesale prices equal the monopoly price since undercutting does not raise market share.

In equilibrium, all home networks invest in steering for efficiency gains, and wholesale prices

decline as a consequence. The game also predicts forming roaming alliances in vertical

pairs, where each pair engages in reciprocal steering to raise the market shares of alliance

members.

Given data before and after the uniform retail pricing policy, the aim of this chapter is to help

us explore the steering behaviours by Etisalat, which affects visited networks market shares.

Although Etisalat owns steering technologies139, we lack prior information on its success and

thus we try to explore the effectiveness of steering in this chapter.

Because steering manipulates the market shares of visited networks, the dependent variable

in this chapter is market share. The model in this chapter does not re-estimate the demand

for visited networks (as in Chapter 6), because it is concerned with the effectiveness of

steering by the home network in response to relative price; henceforth it can consider all

observations.

This chapter is planned as follows. Section (8.1) presents the estimation model and

hypotheses to be tested. Section (8.2) discusses the estimation results. The chapter


138 If the retail price is discriminatory, the rational consumer can select manually between visited networks based on price and non-price characteristics. 139 This statement was communicated verbally by Etisalat.

150


The aim of this chapter is to test the theoretical predictions in light of the impact of Etisalat

uniform retail policy, relaying on the same dataset described in Chapter (5) and used in

Chapters (6) and (7). The dependent variable in this chapter is market shares.

The market shares are calculated without the assumptions made on market size used in

Section (6.2.2). That is, j’s market share equals the quantity of j divided by j’s market

quantities, in each year. We will use fixed and random effect panel regressions for all

observations. The relevant variables used in the estimations are listed in Table (8.1) below.

The market shares variable is a proportion variable distributed on the interval [0,1]. We make

use of the logistic transformation in order to map the variable to the real line. The

transformed shares will remove observations with extreme values (i.e. zero and one), which

is in line with the aim of the model to understand market share allocation within markets. The

market share model for estimation is given by

(

)

( ) ( ) ( )

(8)

where is the error term; and all variables are as listed in Table (8.1).

Notice the use of price relative to the market arithmetic140 mean, instead of the given price.

In Appendix (E), we found that changes in market share (after the uniform retail policy,

assumingly due to Etisalat steering) depend on the relative market price, not on the given

price141. Therefore, the relative price is assumed to matter for relative quantities, from which

market share is derived.

140 Results are very similar if, instead, the geometric mean is used. 141 We tried the estimation with the given price instead of relative price, and found the price coefficients insignificant. In Appendix (E), we re-estimate the demand model as in Eq. (6.1) with the interaction term of price and uniform policy dummy variable. The given price is only significant before the policy for networks operating in highly visited markets, and insignificant (before or after the policy) for networks in low visited markets and for all observations. However, with the relative price instead, price is insignificant (before or after the policy) in the highly visited markets, but significant only after the policy in the low visited markets and in all observations. To conclude, we think the given price suits the demand model as it represents roamers’ response to price (i.e. demand-side substitution), while relative price suits the steering model as it represents Etisalat’s response to price (i.e. supply-side substitution).

151










RPRICEjrt j’s average wholesale price per minute of outgoing call relative to the arithmetic mean of its market r in year t (in constant AED with 2007 as base year).


Description





Description


The use of relative price should help in interpreting the effectiveness of steering. If market

shares respond to changes in relative price during the uniform policy, then steering by

Etisalat is effective, given roamers are not overriding steering with manual network selection.

That is, if Etisalat effectively steers (after the policy), then the coefficient on the interaction of

policy and relative price is expected to be significant and negative. This hypothesis is

explained below.

With discriminatory retail pricing policy, the retail price to roam on network j reflects j’s

wholesale price. If the wholesale price is negatively related to market shares, then roamers

demand-side substitution is effective142. With uniform retail pricing policy, the retail price to

roam on network j does not reflect j’s wholesale price; hence the roamer is indifferent

142 However, demand-side substitution is related to the given price as in Eq. (6.1), not to the relative price.

152

between networks price-wise. In this case, visited networks have no reason to undercut if

steering is absent because undercutting does not raise market shares.

Based on the above, the effectiveness of a home network’s steering should be measured by

how much it is able to replace its roamers’ demand-side substitution with its supply-side

substitution through its ability to steer roamers to the cheapest network. In our case, the

effectiveness of Etisalat steering should be known by how much Etisalat is able to cause

downward pressures on relative wholesale prices during the years when the retail price is

uniform.

The coefficients of the interaction terms related to visited networks’ characteristics are also

relevant in knowing if Etisalat steers to its preferred networks. The hypotheses in this regard

are made in light of Chapter (7) results, where we learned how visited networks charged

Etisalat. Etisalat pays lower wholesale prices to its alliance networks after the policy. Thus,

we expect the coefficient of the interaction of policy and alliance to be positive. On the

contrary, Etisalat gets no wholesale discount from its cross-owned networks. Thus, we

expect the coefficient of the interaction of policy and cross-ownership to be negative.

On the other hand, networks that allow Etisalat prepaid roaming (with CAMEL) are expected

to have higher market shares, regardless of steering, because they affect market size. For

example, Etisalat cannot steer its prepaid roamers to a visited network that does not have

CAMEL technology.

Market structure variables are not included in Eq. (8) because, similar to the demand given

by Eq. (6.1), they are assumed to be correlated with the error term. Nonetheless, Eq. (8)

may still have endogeneity problem because the relative price enters as a regressor and can

be correlated with the error term. We will try instrumenting for it with the use of Nawras IV.


Results for market share estimations are provided in Table (8.2); where the marginal effects

post the uniform policy are presented in Table (8.3). The Hausman test results recommend

the fixed-effect models over the random-effect models. The Chi2 is (1598.59) for models (8a)

153

and (8c); and (61.37) for models (8b) and (8d); which are both significant at 1% level143. We

discuss the results of model (8d) below, with reference to the other models only if they differ

significantly.

As can be seen in Table (8.2), time is insignificant, reflecting no trend in market shares. In

model (8d), the dummy for Etisalat uniform retail policy and the relative price are separately

insignificant, but their interaction term has a negative coefficient which is significant at 1%

level144 (see Table 8.3).

Notice that the policy and the relative price, independently, are significant only if their

interaction term is excluded, (i.e. models 8a and 8c, where their coefficients are negative

and significant at least at 5% level). In other words, the interaction term of those two

variables (i.e. their combination) is what only matters in market shares compared to the

variables independently.

The above results on relative price suggest that only during the policy, a visited network’s

market share is responsive to the wholesale price it chooses145. The question here is why

the choice between visited networks (i.e. market shares) is responsive to wholesale prices

only after the retail price became uniform, where roamers became assumingly indifferent

price-wise. One possible answer is that Etisalat exercised steering. If it did not, the relative

price should remain insignificant. Therefore, we conclude that, after the policy, Etisalat can

effectively steer.

To demonstrate the effect of steering on market shares after the policy, consider the case of

a hypothetical market with three visited networks. Assume the networks are symmetric in

marginal costs and all characteristics. Assume the status quo is such that each network sets

the price per minute at 1 AED.

If network j offers a 10% wholesale discount to Etisalat, j’s relative price is reduced by 7%

(from 1 to 0.93). The marginal effect of such action is 0.052, which is the product of the

change in relative price (i.e. -7%) and the coefficient of the relative price interaction term (-

0.761) in model (8d). By adding 0.052 to the logistically transformed market share (the

143 These results do not change in significance level if ALLIANCEj and CAMELj are excluded from model (8a and 8b) such that all models have the same regressors. 144 This result does not change significantly with or without controlling for visited networks characteristics. 145 This can mean roamers did not care about prices when it was discriminatory. The demand models provided in Appendix (E) show that the relative price is insignificant in highly visited markets, but significant in low visited markets and in all observations only after the policy. However, when the given price is used instead, the given price is significant only before the policy in highly visited markets, but insignificant in low visited markets and in all observations. We conclude that roamers care about prices during the discriminatory period, assuming the given price suits demand estimation because roamers are sensitive to the level of prices.

154

dependent variable in Eq. (8)), we predict that network j will raise its market share by 3.5%.

Table (8.4) demonstrates this hypothetical example.

25Table (8.2) Market share models for all observations.

Dependent variable:

(

)

Random-effect Fixed-effect

(8a) (8b) (8c) (8d)

CONSTANT

-1.197*** -1.834*** -0.818*** -1.366***

(0.207) (0.269) (0.187) (0.251)

YEAR 0.018 0.016 0.008 0.006

(0.050) (0.049) (0.049) (0.049)


-0.364*** 0.498* -0.338*** 0.427

(0.120) (0.263) (0.118) (0.263)

PRICEARjrt (Network own price relative to arithmetic mean)

-0.491*** 0.151 -0.404** 0.144

(0.146) (0.227) (0.161) (0.233)



-2.056*** -1.826*** 1.255 1.423

(0.457) (0.460) (1.168) (1.165)


0.386 0.355 - -

(0.355) (0.354) - -


1.120*** 1.110*** - -

(0.218) (0.217) - -


CROSSOWNEDjt

0.707** 0.506* 1.059*** 0.873***

(0.295) (0.299) (0.295) (0.300)

ALLIANCEj

0.239 0.237 0.264 0.259

(0.169) (0.169) (0.168) (0.167)

CAMELj

0.186* 0.186* 0.141 0.141

(0.108) (0.108) (0.107) (0.107)

PRICEARjrt

- -0.858*** - -0.761***

- (0.233) - (0.234)

Observations 1,706 1,706 1,706 1,706

Visited Networks 499 499 499 499

rho 0.805 0.806 0.860 0.860


155

26Table (8.3) Marginal effects on networks’ characteristics and relative prices post Etisalat uniform retail policy.

Network’s Characteristic And Relative Price

Null Hypothesis Relevant model F-statistics P-value


Random-effect (8b) 10.66 0.001***

Fixed-effect (8d) 3.96 0.047**

Alliance to Etisalat Random-effect (8b) 2.80 0.094*

With CAMEL Random-effect (8b) 36.80 0.000***

PRICEARjrt

Random-effect (8b) 20.37 0.000***

Fixed-effect (8d) 12.69 0.000***

* p<0.10, ** p<0.05, *** p<0.01 - Based on Table (8.2).

27Table (8.4) Hypothetical example of wholesale undercutting in a market with three visited networks.

Relevant variables Status quo Undercut

by 10%

PRICEjt (Network’s own wholesale price)

1.00 0.90

Market arithmetic mean price 1.00 0.97

PRICEARjrt (Network’s own wholesale price relative to the market arithmetic mean)

1.00 0.93

(Network’s own market share) 0.33 0.35

( ) 0.50 0.53

( ( )) -0.69 -0.64

Gain in market share - 3.5%

Turning to visited network’s characteristics, model (8d) suggests that cross-owned networks

are insignificant in the model before the policy; but become significant after the policy. In

fact, after the policy, their market shares are raised at 5% significance level (see Table 8.3).

The result is similar in the other fixed-effect model, (8c).

Similarly, it is obvious that after the policy, the random effect models show an improvement

in market share for the cross-owned networks (e.g. significant at 1% level in model 8b; see

Table 8.3).

156

As found in Chapter (7), cross-owned networks do not offer wholesale discount to Etisalat.

Nevertheless, the results on cross-owned networks after the policy suggest that Etisalat

steers towards its cross-owned networks after the policy. Etisalat probably steers to its

cross-owned networks to help financing them in their expansion stage. There can be other

reasons for which we have no information to control for, related probably to reciprocal IOT

agreements where we observe only one side of it.

However, the contradiction in results between fixed and random effect models is for cross-

owned networks before the policy. The fixed-effect models, before the policy, suggest that

cross-owned dummy variable is insignificant; while the random-effect models suggest it is

significant at 1% level, with negative coefficient. An interpretation is provided below.

We believe there is heterogeneity in visited networks market shares, especially by knowing

that some networks, such as Etisalat cross-owned, inherited very low market shares

because of structural reasons that are unrelated to roaming. For example, cross-owned

networks have either a new license or have very low market share in local subscribers. The

random-effect estimator assumes the intercept is a random variable, which is common to all

networks. Therefore, the heterogeneity problem in market shares magnifies under the

random-effect estimator. This is reflected in the highly significant negative coefficient of the

cross-owned networks (before the policy) in the market share model, which is large in

magnitude.

In contrast, the fixed-effect estimator allows each network to have its own intercept; hence

heterogeneity in market shares is taken into account. This is reflected in the insignificant

coefficient of the cross-owned networks (before the policy) in the market share model.

In addition, the fixed-effect models (8c) and (8d) show the coefficient related to the alliance

dummy variable as insignificant; but we cannot compare the effect to the case before the

policy (i.e. alliance dummy variable is time invariant). In contrast, the random-effect models

(8a) and (8b) show a significant coefficient for the interaction term of the policy and the

alliance dummy variable (at 10% level in model 8b; see Table 8.3). After the policy, the

random-effect models show an improvement in alliance networks’ market shares, which

might be caused by Etisalat steering, given the roamer is indifferent price-wise and alliance

networks significantly charge lower wholesale price as found in Chapter (7).

With regards to visited networks with CAMEL, the fixed-effect models show no impact of this

characteristic on market shares after the policy. In the random-effect models, the policy

157

slightly improves the coefficient’s size of the CAMEL dummy variable. The related

coefficients have the same significance (1%) level before and after the policy (see model 8b

in Table 8.3). This is interpreted as networks with CAMEL can raise market size by hosting

prepaid roamers; and if steering exists (after the policy), steering would be limited to prepaid

roamers and the networks with CAMEL.

Furthermore, similar to the treatment for the price endogeneity problem in the demand

estimation (Section 6.2.4), we tried instrumenting for the relative price used in Eq. (8) with

Nawras IV in the random-effect models (8a and 8b)146. In the 1SLS regressions, Nawras IV

is found insignificant in explaining the relative price. Predictably, Nawras IV does not suit as

an instrument for the relative price147.

In our case, given no other available instruments, it is not feasible to solve the endogeneity

problem when relative price is used as an endogenous regressor. We maintain the relative

price in estimating Eq. (8) because the relative price is assumed to test if market share

allocation (via steering) responds to wholesale price undercutting148.


In this chapter, we modelled market shares to explore for steering, taking advantage of the

dataset on market shares and wholesale prices which are assumed to matter in steering and

roaming alliance formation. Additionally, given uniform retail enables steering, the panel

structure of the dataset makes use of the uniform retail policy to infer about steering.

The existing empirical literature overlooked the importance of the uniform retail in inference

about steering or roaming alliance (see Section 2.3.3). Rieck et al (2005) found that only

Vodafone roaming alliance networks involve in strategic alliance, relying on retail price

information. In the European NRAs competition analyses, the surveys by the NRAs in

Finland and Spain found steering is effective (see Table 2.1).

146 The IV is time invariant; hence it cannot be used with fixed-effect model. Nawras IV interaction with the policy was also tried out for model (8b). 147 This is not surprising, because the IV is supposed to pick up cost differentials that are reflected in prices, but in the case of relative prices, price differentials are minimized (i.e. price is standardised as it is divided by the mean). To confirm, we re-estimated the 1SLS for the demand given by Eq. (6.1) using the relative price instead of the given price. In this case, Nawras IV does not work neither. 148 As explained in Section (8.1), relative price is what matters for relative quantities.

158

In the steering model of this chapter, before the policy (when the retail price is

discriminatory), roamers choice between visited networks does not matter with regards to

the relative price. After the policy (when the retail price is uniform), the choice between

visited networks is significantly sensitive to relative wholesale pricing. We believe steering by

Etisalat is behind this result, as explained below.

With uniform retail price, wholesale undercutting by visited network j does not raise j’s

market share through roamers’ demand-side substitution because the retail price for roaming

on j does not reflect j’s wholesale price. This leaves us with the possibility that the supply-

side substitution (i.e. Etisalat steering) exists and is effective because reduction in relative

wholesale price significantly raises market share.

Steering by Etisalat is found noticeable towards its alliance and cross-owned networks. After

the policy, Etisalat preferred networks (alliance and cross-owned) experienced

improvements in their market shares, particularly the cross-owned networks.

159

Chapter (9). Conclusion

Mobile international roaming is a hot topic in the telecommunications industry, particularity in

the EU where international roaming raised competition concerns that led to price regulation.

The NRAs that studied their wholesale market concluded that the market was competitive;

however, the EC was concerned that reductions in wholesale prices were not passed on to

consumers. The EC was also concerned that prices (retail and wholesale) remained

unjustifiably high with no sign of decline.

A possible reason for the competition concerns is the high consumers’ switching costs that

made retail prices unjustifiably high. Sources of switching costs include the inconvenience to

use the outside options (e.g. public payphones or visitor-SIM card), the absence of mobile

number portability at the retail market, and probably the roamers’ temporary nature of visit.

However, the European NRAs competition analyses and the theoretical literature mainly

focused on the wholesale market, which is also the focus of this thesis.

This thesis studies the economics of mobile international roaming with emphasis on visited

networks’ wholesale competition in hosting the visiting subscriber of their IOT-agreement

partners. Areas tackled in the thesis are similar to the ones tackled by the existing theoretical

literature. In addition, the thesis uses a dataset on Etisalat outbound roaming to test the

theoretical predictions.

All the existing published papers on the economics of international roaming assume uniform

retail price. This thesis demonstrates that using such assumption in visited networks’

optimization problems leads to results that are inconsistent with wholesale competition, for

example the number of visited networks increases wholesale price. The source of such

inconsistency is the fact that visited networks appear in the demand as (perfect)

complements.

We compare equilibria under the two different assumptions on retail pricing policy (uniform

and discriminatory). As found in Proposition (3.1), the discriminatory retail price suits for

modelling wholesale competition, because visited networks appear in the demand as

substitutes and price equilibria are in accordance with competition models.

Wholesale competition is found in our empirical estimations on the effect of market structure

on wholesale price (Chapter 7). We found that the number of visited networks, as a measure

160

of wholesale competition, significantly reduces wholesale prices. In addition, horizontal

ownership (e.g. merger) enables visited networks to raise wholesale prices.

In addition, our base model predicts that a nation’s (unweighted) welfare can be enhanced

by (retail and wholesale) price reduction, supporting the argument for price regulation.

Nonetheless, because of the cross-border nature of the IOT agreements, price regulation

requires the cooperation of the relevant NRAs.

Another policy implication is the use of a structural model for demand estimation (Chapter 6),

similar to our simple logit model, to derive the own price elasticity that is crucial in industry’s

impact assessment analyses on the EC roaming regulation. A future industry assessment

can consider a European home network, say Vodafone-UK, to study quantities and prices149

related to its outbound roamers within the EU. Of course, including information on Vodafone-

UK’s retail pricing policy (discriminatory or uniform) and information on Vodafone-UK’s

relationship with visited networks (e.g. cross-owned or roaming alliance) will help in

estimating the demands for visited networks.

Our market size calculation can be used in future demand estimation, but with better

information on, for example, the demographics of Vodafone-UK’s roamers in each visited

market. Better information can allow for sophisticated demand estimation models, such as

the random coefficients logit.

The instrumental variable method used in our demand estimation can be suited for future

research. In our experiment, we found that uniform retail prices are better instruments

compared to discriminatory prices, which is interpreted as uniform prices overcoming the

bias in wholesale preferential agreements.

The two assumptions on the retail policy (discriminatory or uniform) lead to opposite results

with regards to the incentive for vertical cross-ownership. At one hand, with discriminatory

policy, the incentive for cross-ownership exists (Proposition 4.1). But with the uniform policy

on the other hand, such incentive does not exist (Proposition 4.2). In the cross-ownership

games (either with discriminatory or uniform retail policy), it is predicted that vertically cross-

owned networks charge internally preferential wholesale prices.

149 Data on wholesale prices charged to Vodafone-UK may better suit the demand estimation for visited networks compared to Vodafone-UK’s retail prices if those retail prices are based on zones (e.g. a price to roam in EU and a price to roam outside EU) such that they lack enough variation for analysis.

161

However, empirically (Chapter 7), we found that Etisalat cross-owned networks do not offer

preferential wholesale prices to Etisalat. We looked for the nature of cross-owned networks

to seek an interpretation. All of these networks operate in developing countries, where

penetration is relatively low; and many of them are new in their markets or/and have low

market shares in local subscription. Therefore, these networks may be in the expansion

phase; hence behaving non-cooperatively.

After Etisalat uniform retail policy, Etisalat roamers are found to have negative mean utility

by roaming on cross-owned networks, probably due to their low-quality of coverage.

Paradoxically, after the policy, cross-owned networks significantly improved their market

shares. This suggests that Etisalat exercise steering towards its cross-owned networks.

Given data limitation, we do not know why Etisalat would steer towards its cross-owned

networks despite, in return, not getting wholesale discount.

By Proposition (4.3), all home networks are predicted to invest in steering technology for

efficiency gains when substitutability is high enough to drive down wholesale prices. The

roaming alliance model predicts that mutual steering between allied networks can be self-

sustained given no incentive to deviate. Proposition (4.4) predicts that networks will engage

in roaming alliance pairing, because joining a roaming alliance is a dominant strategy to

each network. The equilibrium is a prisoner dilemma situation where the status quo profit is

reduced by the sunk cost of steering investment.

In exploration for steering in our dataset (Chapter 8), we conclude that steering by Etisalat is

effective because, after its uniform retail policy (where roamers are assumed indifferent

price-wise), visited networks significantly reduced wholesale prices to gain market share.

Additionally, Etisalat roaming alliance networks are found to offer lower wholesale prices to

Etisalat only after the policy (Chapter 7). The improvement in their market shares after the

policy is interpreted as a result of steering by Etisalat.

All the equilibria in the games of preferential wholesale agreements do not lead to exclusive

IOT agreements, because substitutability of visited networks is assumed to be bounded.

This is reflected in the dataset, where Etisalat increases its IOT-agreement partners every

year during the study period in spite of the existence (and the rise) of preferred visited

networks.

Given the role of net payments in all the games of the preferential wholesale agreements,

we expect large networks (e.g. in market size) would prefer to be preferential partners to

162

minimize negative net payments. To understand the choice of preferential partner, an

extension for future work is to consider asymmetry in networks in terms of market size,

marginal cost, and the number of networks in a country.

We ignored network-coverage decisions, relying on assuming such decisions are only

relevant for home subscribers, not for hosting temporary roamers. Future work may consider

an investment game to improve coverage in order to capture visitors in areas such as

airports, learning from the model developed by Ambjørnsen et al (2011).

Additional information on visited networks should improve our empirical estimations. The age

of mobile networks can help control for networks in their early phase of expansion, in which

case, those networks can rely on unregulated wholesale prices (e.g. IOTs) as interpreted for

Etisalat cross-owned networks. Moreover, prices before the year 2007 can help test if there

is waterbed effect caused by the EC roaming regulation. In addition, a visited network

market share in local subscribers should be informative about its market share in roaming. If

the network is dominant in the local market, it should have higher coverage or quality and

hence higher share in hosting visitors.

All empirical models in this thesis consider quantities and wholesale prices related to the

service of roaming outgoing calls, because among roaming services, the service of outgoing

calls is assumed to be exposed to the effects of the shift in Etisalat retail policy. The service

of outgoing calls was also the focus of the NRAs’ analyses in the EU for the wholesale

market of international roaming.

We expect to have similar results if market share estimation is made for the other roaming

services separately (i.e. incoming calls, SMS and data roaming). This is so because market

shares of those independent services are found highly correlated due to demand-side or

supply-side substitution (via home network steering) related to price, or non-price reasons

(e.g. the quality of network-coverage).

Future research may employ a model that comprehends all the roaming services. Retail

roaming packages (e.g. Etisalat’s incoming calls and data roaming) should also be

considered in future research. Instrumental variables can be used for each roaming service

separately (e.g. Nawras’ retail prices for SMS as IV for SMS wholesale prices charged to

Etisalat).

163

Nevertheless, some data problems may arise when expanding estimations to the other

roaming services. As seen in our dataset with incoming calls, wholesale prices would include

many observations with zero or near zero prices, because visited networks get compensated

from international settlement rates. Moreover, with data roaming, the roamer may face

difficulty in measuring his usage150.

The implications of the net payment for the IOT-agreement partners are deemed to be the

cause of all the incentives in the games of cross-ownership, steering and roaming alliance

formation. However, our dataset is about Etisalat outbound roaming, i.e. Etisalat’s IOT out-

payments. If data on inbound roaming (i.e. foreign subscribers roaming on Etisalat network

inside the UAE) is available, we would have Etisalat’s IOT in-payments. Information on the

two sides of IOT net payments should enable investigating the determinants of IOTs (e.g.

the level of IOTs), learning from the empirical model by Wright (1999). Additionally,

information on the retail policy of Etisalat IOT-agreement partners is relevant, especially

whether Etisalat is considered a roaming alliance at the side of those partners. Information

on the two sides of an IOT net payment should enable testing for steering reciprocity in order

to understand the behaviour of roaming alliances, as approached by Rieck et al (2005).

Furthermore, the observed wholesale prices (IOTs) charged by the visited networks to

Etisalat may be affected by Etisalat’s non-roaming interconnection agreements with those

visited networks, such as the agreement of international settlement rates. Despite the

independence of the IOT and the international settlement rate agreements, the bargaining

positions of the parties involved can affect any wholesale price they choose.

The theoretical models in this thesis assume monopoly retail, which is thought to be

plausible if the market does not have mobile number portability, as with our case in the UAE.

However, many markets have mobile number portability, where retail competition for

subscribers may influence international roaming retail offering. A good example is the market

in Saudi Arabia where mobile number portability exists. The new entrant Zain started offering

free incoming calls for roamers to gain market share in subscribers. In response to the

churn-rate caused by Zain, the other networks (STC and Mobily) matched Zain’s offer. The

Saudi NRA considered free incoming calls as below-cost pricing (Sutherland, 2011). Future

research should consider retail offering in modelling mobile networks that compete at the

retail level in international roaming.

150 According to NRA-UAE, most of consumer complaints are about their bill shocks for international roaming, in particular, data roaming.

164

The thesis, theoretically and empirically, focuses on the demands for visited networks to

understand wholesale competition. An interesting future work is the demand for visited

markets. There are hot destinations that attract UAE outbound tourists (where Etisalat

roamers can serve as a proxy), such as neighbouring GCC countries and the UK. With a

random-effect regression estimator, we tried out regressing the log of market quantity on

time, the distance between the UAE and the visited market, GDP of the visited market, and

the total trades between the UAE and the visited market. All these explanatory variables are

highly significant except total trades: time and distance have negative coefficients; and GDP

has positive coefficient. However, the coefficients are very small in magnitude.

In the future, we plan to do a separate research on the demand for markets, by including

additional information such as the number of flights and the foreign immigrants in the UAE.

We plan to shade light on the low visited markets, which seem to host merely postpaid

roamers. If those markets are mainly visited by the diplomats for example, one can consider

including additional information on the number of staff in the UAE embassies.

An observed business practise in this industry is the emergence of Inbound Roaming

Promotion by visited networks (e.g. Viva-Kuwait and Batelco-Bahrain). The promotion is

usually a draw to win a prize for roaming on a visited network. Visited networks inform about

these prizes through advertisements contained in the welcoming SMSs sent to roamers. A

model about the effect of such advertisement on demand for a visited network is awaiting

future research.

165

Bibliography Ambjørnsen, T., et al. (2011). "Customer ignorance, price-cap regulation, and rent-seeking in mobile roaming."

Information Economics and Policy 23(1): 27-36.

ARCEP (2006). The Market for International Roaming. Public consultation (January 10th -February 21st 2006),

Autorité de Régulation des Communications Électroniques et des Postes (ARCEP) January 2006.

Armstrong, M. (2002). The Theory of Access Pricing and Interconnection. In: M. Cave, S. Majumdar, and I.

Vogelsang (eds.), Handbook of Telecommunications Economics. Netherlands, Elsevier.

Armstrong, M. and J. Wright (2009). Mobile Call Termination in the UK: A Competitive Bottleneck? In: B. Lyons

(ed.), Cases in European Competition Policy: The Economic Analysis USA, Cambridge University Press.

Belleflamme, P. and M. Peitz (2010). Product Differentiation. In: P. Belleflamme and M. Peitz, Industrial

Organization: Markets and Strategies. USA, Cambridge University Press.

Berry, S. (1994). "Estimating Discrete-Choice Models of Product Differentiation." The RAND Journal of Economics:

242-262.

Berry, S., et al. (1995). "Automobile Prices in Market Equilibrium." Econometrica: Journal of the Econometric

Society: 841-890.

Bonanno, G. and J. Vickers (1988). "Vertical Separation." The Journal of Industrial Economics 36(3): 257-265.

Buigues, P. A. and P. Rey (2004). The Economics of Antitrust and Regulation in Telecommunications: Perspectives

for the New European Regulatory Framework. UK, Edward Elgar Publishing.

Carter, M. and J. Wright (1994). "Symbiotic Production: The Case of Telecommunication Pricing." Review of

Industrial Organization 9(4): 365-378.

Cave, M. and M. P. Donnelly (1996). "The Pricing of International Telecommunications Services by Monopoly

Operators." Information Economics and Policy 8(2): 107-123.

CMT (2009). Report on the Analysis of the International Roaming Service in the Spanish Mobile Telephone Market,

Comisión del Mercado de las Telecomunicaciones.

Cullen-International (2013). Summary Reports on International Roaming.

Cyrenne, P. (1994). "Vertical Integration versus Vertical Separation: An Equilibrium Model." Review of Industrial

Organization 9(3): 311-322.

Dam, R. (2011). "Du: The Next Level." Win-Win/ Jun 2011. Retrieved 18 November, 2013, from

http://www.huawei.com/en/static/hw-093254.pdf.

Dixit, A. (1979). "A Model of Duopoly Suggesting a Theory of Entry Barriers." The Bell Journal of Economics 10(1):

20-32.

EC (2000). Working Document on the Initial Findings of the Sector Inquiry into Mobile Roaming Charges.,

European Commission.

EC (2006). Impact Assessment of Policy Options in Relation to a Commission Proposal for a Regulation of the

European Parliament and of the Council on Roaming on Public Mobile Networks Within the Community. SEC

(2006), European Commission. 926.

166

Economides, N. (1994). The Incentive for Vertical Integration. Working Papers, Stern School of Business, N.Y.U.

Einhorn, M. A. (2002). "International Telephony: A Review of the Literature." Information Economics and Policy

14(1): 51-73.

EITC (2008). "Du Introduces First Unified Roaming Tariff in the GCC ". Retrieved 11 November, 2013, from

http://www.du.ae/en/about/media-centre/newsdetails/du-introduces-first-unified-roaming-tariff-in-the-GCC.

Etisalat (2009). PostPaid Mobile Roaming Charges For Etisalat Subscribers as of August 2009, Etisalat. 13 August

2009.

Etisalat (2010a). "Etisalat International Roaming Partners List as of July 2010." Retrieved 18 November, 2013, from

http://www.etisalat.ae/ar/system/docs/personal/misc/Roaminglist.pdf.

Etisalat (2010b). "PostPaid Mobile Roaming Charges For Etisalat Subscribers as of November 2010." Retrieved 18

November 2013, from http://www.etisalat.ae/en/system/docs/personal/misc/PostPaidRoamingCharges.pdf.

Etisalat (2010c). "PrePaid Mobile Roaming Charges For Etisalat Subscribers as of November 2010." Retrieved 18

November 2013, from http://www.etisalat.ae/en/system/docs/personal/misc/PrePaidRoamingCharges.pdf.

Etisalat (2011). Annual Report 2011. Annual Reports. . http://www.etisalat.com/en/system/docs/reports/etisalat-

2011-en.pdf.

Etisalat (2013a). Etisalat International Roaming Partners List as of May 2013. 2 September 2013.

Etisalat (2013b). "Etisalat Preferred Roaming Partners List as of July 2013." Retrieved 18 November, 2013, from

http://www.etisalat.ae/en/system/docs/personal/misc/PreferredRoaminglist.pdf.

European-Parliament (2006). Technical Issues On Roaming: Transparency, Technical Aspects and Data – Overview

Related to the Proposed Regulation on Roaming. Briefing Note (IP/A/ITRE/FWC/2006-087/Lot2/SC1), European

Parliament.

Europe-Economics (2008). Review of the Roaming Regulation. Study was requested by the European Parliament's

committee on Internal Market and Consumer Protection (IMCO). IP/A/IMCO/FWC/2006-186 C, Europe Economics.

Fabrizi, S. and B. Wertlen (2008). "Roaming in the Mobile Internet." Telecommunications Policy 32(1): 50-61.

Falch, M., et al. (2009). "International roaming: is there a need for EU-regulation beyond 2010?" info 11(4): 19-33.

Gans, J. S. and S. P. King (2000). "Mobile Network Competition, Customer Ignorance and Fixed-To-Mobile Call

Prices." Information Economics and Policy 12(4): 301-327.

GSMA (2008). GSM Association Briefing Paper On the Impact Assessment Accompanying the Proposal for a

Regulation of the European Parliament and of The Council Amending Regulation (EC) No 717/2007 on Roaming

on Public Mobile Phone Networks Within the Community, GSMA.

GSMA (2013). "About Roaming." Retrieved 15 November, 2013, from http://www.gsma.com/aboutus/gsm-

technology/roaming.

Gulfnews.com (2009). "Etisalat Expands Global Footprint ". Retrieved 11 November, 2013, from

http://gulfnews.com/business/telecoms/etisalat-expands-global-footprint-1.46293.

Gulfnews.com (2010). "Etisalat Simplifies Global Roaming Rates." Retrieved 11 November, 2013, from

http://gulfnews.com/business/telecoms/etisalat-simplifies-global-roaming-rates-1.580219.

Hakim, S. R. and D. Lu (1993). "Monopolistic Settlement Agreements in International Telecommunications."


167

Hausman, J., et al. (1994). "Competitive Analysis with Differenciated Products." Annales d'Economie et de

Statistique: 159-180.

Hoernig, S. and T. Valletti (2012). Mobile Telephony. In: M. Peitz, and J. Waldfogel (eds.), The Oxford Handbook of

the Digital Economy. USA, Oxford University Press.

Hualong, Y. and T. Shoulian (2009). A Study of the Roaming Alliance in Mobile International Roaming Market.

New Trends in Information and Service Science, 2009, IEEE.

Lacasa, J. D. (2011). Competition for Partners: Strategic Games in Wholesale International Roaming. 22nd

European Regional ITS Conference, Budapest 2011.

Laffont, J. J., et al. (2000). Multiple Bottlenecks and Two-Way Access. In: J. Tirole and J.J. Laffont, Competition in

Telecommunications. UK, MIT Press.

Lupi, P. and F. M. Manenti (2009). "Traffic Management in Wholesale International Roaming: Towards a More

Efficient Market?" Bulletin of Economic Research 61(4): 379-407.

Majumdar, S. K. (2010). "Institutional Changes, Firm Size and Wages in the Telecommunications Sector."


Majumdar, S. K., et al. (2012). "Mergers and Synergy: Lessons from Contemporary Telecommunications History."

Telecommunications Policy 36(2): 140-154.

Martino, M. (2007). "Effectiveness of Traffic Direction Techniques in Mobile International Roaming and

Implications for Market Definition." COMMUNICATIONS AND STRATEGIES 66: 137.

NBS (2013). Consumer Price Indices 2008-2011. UAE, National Bureau of Statistics in the UAE.

N " u g k w ‐ ‐ u " Econometrica 69(2): 307-342.

NRA-Oman (2013). Nawras Retail Prices as of August 2013, Telecommunications Regulatory Authority of the

Sultanate of Oman.

NRA-UAE (2007). International Roaming Notification Directive. UAE, Telecommunications Regulatory Authority of

the UAE. Directive No. (3) of 2007.

NRA-UAE (2010). Intra-GCC Roaming Charges Directive. UAE, Telecommunications Regulatory Authority of the

UAE. Directive No. (4) Of 2010.

O.J. (2001). Notification of Cooperation Agreements. C42, Official Journal of the European Communities.

O.J. (2003). Commission Recommendation on Relevant Markets. L 114, Official Journal of the European

Communities.

Oman-Mobile (2013). "Mada Roaming Charges." Retrieved 21 October, 2013, from

http://www.omantel.om/OmanWebLib/Travellers/Abroad/Mobile/Roaming%20Charges.aspx?linkId=4&MenuId=1

67.

Peitz, M., et al. (2004). "Competition in Telecommunications: An Introduction." Information Economics and Policy

16(3): 315-321.

Rieck, O., et al. (2005). Alliances of Networks and Networks of Alliances: International Cooperation in Mobile

Telecommunications. 16th European Regional ITS Conference. Porto, Portugal, 2005.

Salsas, R. and C. Koboldt (2004). "Roaming Free? Roaming Network Selection and Inter-Operator Tariffs."


168

Shortall, T. (2010). A Structural Solution to Roaming in Europe. Florence School of Regulation.

Shubik, M. and R. Levitan (1980). Duopoly with Product Differentiation. In: M. Shubik with R. Levitan, Market

Structure and Behaviour. USA, Harvard University Press.

Shy, O. (2001). Telecommunication. In: O. Shy, The Economics of Network Industries. UK, Cambridge University

Press.

Singh, N. and X. Vives (1984). "Price and Quantity Competition in a Differentiated Duopoly." The RAND Journal of

Economics 15(4): 546-554.

Slade, M. E. (2009). Merger Simulations of Unilateral Effects: What Can We Learn From the UK Brewing Industry?

In: B. Lyons (ed.), Cases in European Competition Policy: The Economic Analysis USA, Cambridge University Press.

Sonnenschein, H. (1968). "The Dual of Duopoly is Complementary Monopoly: Or, Two of Cournot's Theories are

One." The Journal of Political Economy 76(2): 316-318.

Spengler, J. J. (1950). "Vertical Integration and Antitrust Policy." The Journal of Political Economy 58(4): 347-352.

Sutherland, E. (2008). "The Regulation of International Mobile Roaming." info 10(1): 13-24.

Sutherland, E. (2011). "International Mobile Roaming in the Arab States." Information Economics and Policy 13 (2).

Tsyganok, L. (2008). Economic Analysis of International Mobile Telecommunication Services: The EU Roaming

Markets Case, Discussion paper, mimeo.

Valletti, T. (1999). "A Model of Competition in Mobile Communications." Information Economics and Policy 11(1):

61-72.

Valletti, T. (2003). "Is Mobile Telephony a Natural Oligopoly?" Review of Industrial Organization 22(1): 47-65.

Valletti, T. (2004). "Obligations that Can be Imposed on Operators with Significant Market Power under the New

Regulatory Framework for Electronic Communications." RICHARD A. CAWLEY 3 5(1): 51.

Vickers, J. and M. Waterson (1991). "Vertical Relationships: An Introduction." The Journal of Industrial Economics

39(5): 445-450.

Wright, J. (1999). "International Telecommunications, Settlement Rates, and the FCC." Journal of Regulatory

Economics 15(3): 267-292.

169

Appendix (A). Choice On Retail Policy By A

Monopolist

In this appendix, we focus on the problem of a monopolist, ignoring any vertically related

issues, in the choice of (linear) retail pricing policy.

Consider a monopolist selling two substitutable goods that have different actual marginal

costs. The monopolist’s problem is to choose a retail pricing policy. That is, the monopolist

either chooses discriminatory prices (i.e. different prices based on marginal costs) or a

uniform price (i.e. common price regardless of marginal costs). The demand for good is

given by Shubik and Levitan (1980)

( .

/

* ( ) ( )

If price is uniform, then the demand for each good is identical

( )

Next, equilibria are presented for the discriminatory and the uniform policies.

A1. Discriminatory Prices (Policy D)

Under the discriminatory pricing policy (Policy D), the maximization problem is given by

( )

( ) ( ) ( ) ( )

The first order condition (FOC) for the above maximization problem leads to the following

retail solution for good .

Therefore, the equilibrium profit from selling good is

( )[ ( ) ( )]

( ) ( )

and the total profit from selling both goods is

,( ) ( )

- ( )

(A.1)

A2. Uniform Price (Policy U)

Under the uniform pricing policy (Policy U), the maximization problem is given by

170

( )

( ) ( )

where ( ) , which is the average marginal cost. The FOC for the above

maximization problem leads to the following uniform retail solution.

Therefore, the equilibrium profit from selling good is

( )

( )

and the total profit from selling both goods is

( )

( ( ))

(A.2)

A3. Payoffs Comparison

This section compares the payoffs from the discriminatory and uniform pricing policies. The

difference between Eq. (A.1) and Eq. (A.2) is given by

( )( )

(A.3)

It is clear from Eq. (A.3) that if marginal costs are symmetric, the monopolist is indifferent

between discriminatory and uniform pricing policies. This is true because the prices of the

two goods would be symmetric. However, with asymmetric marginal costs, the monopolist is

better off choosing the discriminatory policy than choosing the uniform. This conclusion is

more appealing as the two goods become more asymmetric in marginal costs or/and more

substitutable.

171

Appendix (B). Derivations Of Base Model Equilibria

This Appendix shows the derivations of the base model equilibria presented in Chapter (3).

B1. Model Background

There are two networks in country ( and ); and two networks in country ( and

). Each network signs an IOT agreement with each foreign network. Each network is

assumed to be a monopolist at the retail level and a duopolist at the wholesale level.

Networks are assumed to have symmetric constant marginal costs normalized to zero and

no fixed costs. Retail and wholesale prices are assumed to be linear.

The game is played in two stages: in Stage I, networks set their wholesale prices

simultaneously; and in Stage II, networks set their retail prices simultaneously. The subgame

perfect Nash equilibrium (SPNE) is solved by backward induction.

To simplify notations, we solve for the situation ’s subscriber roaming on ’s network

under the ( ) IOT agreement. We show the retail price solution for and the

wholesale price solution for . In such two-way access agreements, the solutions apply

similarly the other way to all networks. The subscript refers to the network in question and

the superscript refers to its IOT-agreement partner.

The utility for ’s subscriber is given by

(

)

( )

(B.1)

where

is the aggregate quantity demanded by ’s subscriber in country ;

are the applicable retail prices; and are positive demand parameters; and

is the substitutability parameter. We will later write ( ) to index the substitutability

level in each country, where , ).

The roamer maximizes his utility for roaming in country by choosing and

(B.2)

The direct demand for roaming on networks and respectively are given by

172

[ .

/

]

and

[ .

/

]

(B.3)

If the prices are uniform (i.e.

), the demands will be symmetric

, -

(B.4)

The different wholesale pricing strategies explored in the models are considered under two

alternative retail-pricing policies: with uniform constraint or without uniform constraint (i.e.

discriminatory by visited networks). Such policies produce different retail price mapping

functions (RPMFs), which are derived in Stage II. A RPMF is a function of wholesale

price(s), which provides the optimal retail price given any previously set wholesale price(s),

and is substituted for when solving for wholesale prices in Stage I151.

B2. The Game

B2.1 Stage II

Discriminatory Retail Policy

Home network faces the following profit maximization problem when its subscriber roams

in country

(

)

(

) (

)

(

)

(

) (

)

(

)

(B.5)

’s retail prices are inside its retail profits (the first and the third terms, where the demands

are given by Eq. (B.3)). The RPMFs for both networks are derived by solving

simultaneously, where

(

) (

)

(

)

(

)

(

)

151 The second order condition is negative in all retail (discriminatory or uniform) and wholesale maximization problems.

173

[ .

/

]

.

/ (

)

(

)

and similarly for ; yielding the following retail price functions

( ) (

)

( )

and

( ) (

)

( )

By solving for , we have

( )

, ( )

-

( )

( ) ( ) ,( ) -

( )( )

where ( ). After substituting for in the above two equations, the last term in the

numerator becomes zero. Then by multiplying the numerator and the denominator by (

), the RPMF is given by

and similarly for

(B.6)

Uniform Retail Policy

In the absence of steering and with uniform retail, home network sets identical retail

prices in country , resulting in identical demands. The average wholesale price faces is

(

) . The following profit maximization problem is for when its

subscriber roams in country

( )

( ) ( )

( ) ( )

(B.7)

’s retail price is inside its retail profit (the first term, where the demands are given by Eq.

(B.4)). The first order condition (FOC) must satisfy

( ) (

) * ( )

+

, -

(

)

By solving for , the RPMF is given by

174

(B.8)

Note importantly that, as compared to Eq. (B.6), depends on both and

.

Next, four wholesale pricing strategies are presented. In each wholesale strategy, networks

are assumed to set their (monopoly) retail prices independently according to the relevant

RPMF. Any price cooperation between networks is only assumed upstream. The following

first order conditions (FOCs) are with respect to (i.e. ’s wholesale price to ), where

’s retail price is given by the RPMFs of Eq. (B.6) for the discriminatory retail policy, or by

the RPMF of Eq. (B.8) for the uniform retail policy.

B2.2 Stage I

B2.2i International Cartel Wholesale Pricing Strategy

This strategy sets wholesale prices to maximize the following international market profit

(

)

( )

(

)

(

)

(

)

(

)

(

)

(

)

(

)

(

)

(B.9)


Using the demands as in Eq. (B.3), and substituting Eq. (B.6) in Eq. (B.9), we find a


( )

(

) *

(

)

+

*

(

)

+

[ .

/

]

.

/

( )

where ( ). Substituting further the relevant RPMFs (from Eq. B.6) in the above

equation, we have

(

)

( )

175

By solving for , we have

[

]

The second term cancels in the above equation after substituting for . The international

cartel’s best wholesale price is given by

(B.10)

Because , ) , the only symmetric wholesale equilibrium is when , otherwise

. Thus the equilibrium wholesale and retail solutions are

then from Eq. (B.6)

(B.11)




( )

( ) *

( )

+

, -

By substituting further for the RPMF (from Eq. B.8) in the above equation, we have

(

*


(B.12)

The only symmetric wholesale price solution is zero. Thus the equilibrium wholesale and

retail solutions are

then from Eq. (B.8)

(B.13)

176

B2.2ii National Cartel Wholesale Pricing Strategy

This strategy sets wholesale prices to maximize the national market profit. The maximization

problem for networks in country is

(

)

(

) (

)

(

)

(

) (

)

(

)

(

) (

)

(

)

(

) (

)

(

)

(B.14)




(

) *

(

)

+

*

(

)

+

[ .

/

]

.

/

By substituting further for the relevant RPMFs (from Eq. B.6) in the above equation, we have

* .

/(

)

(

)+

.

/


( )

(B.15)

where ( ) . By substituting symmetrically for in the above equation, then

substituting for , we get

( )( )

then from Eq. (B.6),

(

*

(B.16)

177




( )

* ( )

+

* ( )

+

, -


[ (

(

*

,]


(B.17)

The symmetric wholesale price requires

; thus Eq. (B.17) gives us

Therefore, the equilibrium wholesale price solution that maximizes the national market profit,

and its resulting retail price solution are


(

*

(B.18)

B2.2iii Narrow Vertical Wholesale Pricing Strategy

This strategy sets wholesale prices to maximize the vertical retail revenues of the two IOT

partners within each separate IOT agreement (i.e. ignoring implications for other revenues

received by the networks). The maximization problem for the networks in the ( ) IOT

agreement is

(

)

( )

(

)

(

)

(B.19)

178

The above equation will result in symmetric wholesale prices for all networks, in line with the

rest of the strategies of this chapter.




( )

(

) *

(

)

+

[ .

/

]

.

/

( )

where ( ). By substituting further for the relevant RPMFs (from Eq. B.6) in the

above equation, we have

(

)

( )

By solving for , the vertical strategy’s best wholesale price is given by

(B.20)

By substituting symmetrically for in the above equation, we get

( )


(

( )

*

(

*

(

*

( )

(B.21)

Since , ), if . The negative wholesale price reduces the retail price and

effectively enhances the demand for the IOT partner. Intuitively, the mutual wholesale price

subsidy is necessary to bring down the retail price.




179

( )

( ) *

( )

+

, -


(

*

By solving for , the vertical strategy’s best wholesale price is given by

(B.22)

The only symmetric wholesale price solution is zero. Thus the equilibrium wholesale and

retail solutions are


(B.23)

B2.2iv Unilateral Wholesale Pricing Strategy

This strategy sets wholesale prices to maximize the own wholesale profit for each network.

The maximization problem for is

(

)

(

) (

)

(

)

(

) (

)

(

)

(B.24)




(

) *

(

)

+

[ .

/

]

.

/

By substituting further for the relevant RPMFs (from Eq. B.6) in the above equation, we have

* .

/(

)

(

)+

.

/

180


( )

(B.25)

where ( ) . By substituting symmetrically for in the above equation, then

substituting for , we get

( )( )

( )


(

( )

*

( )

( )

(B.26)




( )

* ( )

+

, -


[ (

(

*

,]


(B.27)

By substituting symmetrically for in the above equation, we get


(

*

(B.28)

181

Appendix (C). Networks Choice On Retail Policy

In this Appendix, we show a game whereby mobile networks choose wholesale prices

simultaneously, then decide on a retail policy (either discriminatory or uniform at-the-

monopoly-level).

C1. Model Background

Assume there are two countries; each has two networks. Each network signs an IOT

agreement with each foreign network. Each network is a monopolist downstream and a

duopolist upstream. Networks have symmetric constant marginal costs normalized to zero,

no fixed costs, and no steering technology. Retail and wholesale prices are linear.

The one-shot pricing game is played in two stages. In Stage I, networks simultaneously

choose wholesale prices. In Stage II, networks simultaneously either choose retail prices

discriminated by visited networks (hereafter discriminatory policy) or choose a uniform retail

price that equals the monopoly price (hereafter uniform policy). The subgame perfect Nash

equilibrium (SPNE) is solved by backward induction.

The demand for visited network is given by Shubik and Levitan (1980)

[ .

/

] ( ) ( )

Networks have symmetric demand parameters ( ), with and (the

substitutability parameter). If the prices are identical (i.e. ), the demands will be

symmetric,

, -

Consider the unilateral equilibria of the discriminatory retail pricing policy of the base model

in Section (3.2.8iv). When wholesalers set wholesale prices non-cooperatively, the

equilibrium wholesale and retail prices are, respectively, , ( )- and

( ) , ( )- . Therefore, the demand for a visited network is ( )

, ( )-.

Consider the case the home network chooses the monopoly price, which equals

( ). The demand for a visited network, therefore, is . Given uniform retail price

and absence of steering, wholesalers do not have an incentive to undercut because

182

undercutting does not raise market shares. In this case, the wholesale price will equal the

monopoly price, i.e. ( ).

In the following game, we will be comparing profits, where the profit to each network from

each IOT agreement consists of a retail revenue and a net payment. The net payment is the

difference between a network’s wholesale profit and wholesale cost with its IOT-agreement

partner. Each term in the net payment (either wholesale profit or wholesale cost) consists of

a wholesale price times a demand.

In our case, the wholesale profit/cost to any network is either or . Notice that

( ) , ( )- as long as ; otherwise . Notice also that

( ) , ( )- except in the limit of , where , (because the

respective retail prices would be equal).

Assume network , where is any network, chooses the discriminatory policy while its IOT-

agreement partner chooses the uniform policy. Then, the net payment to network is

⏞

⏞

( )

⏞

( )

( )

⏞

( )

( )

(C.1)

Network receives the net payment from its IOT partner because charges a higher

wholesale price (i.e. ) and has a lower (retail) demand (i.e. ).

C2. The Game

The total profits, , to any network from its both IOT agreements if all networks apply the

same policy will equal its total revenues, . This is because net payments are zero in this

case.

When all networks apply the uniform policy, the total profits/revenues to each network are

(C.2)

When all networks apply the discriminatory policy, the total profits/revenues to each network

are

183

( )( )

( )

(C.3)

The difference between Eq. (C.2) and (C.3) is positive,

( )( )

( )

( )

(C.4)

Therefore, all networks are better off if they all choose the uniform policy (with retail price at-

the-monopoly-level). However, there is an incentive to deviate from such cooperation, as

explained next.

If network plays the discriminatory policy, and only one of its IOT-agreement partners plays

the uniform policy, ’s total payoffs are

( )

( )

(C.5)

where is given by Eq. (C.1).

The difference between Eq. (C.5) and (C.2) is

( )

(C.6)

which is positive only if . If, instead, both IOT partners of network play uniform policy,

network will have two net payments; hence Eq. (C.5) will increase to .

Therefore, each network has an incentive to deviate from cooperation in choosing uniform

policy by playing the discriminatory policy. As a result, discriminatory policy is a dominant

strategy to each network, which is the equilibrium in this one-shot game as long as visited

networks are substitutes (i.e. ).

C3. Discussion

In this one-shot game, the equilibrium payoffs to each network is , which is inefficient

because it is less than when all networks choose the uniform policy (i.e. retail price at-

the-monopoly-level). In a repeated game with infinite time horizon, networks can sustain

setting the retail price at the monopoly level. Because steering is assumed absent, the

uniform retail price provides the wholesaler with no incentive to undercut, as undercutting

184

does not increase market share. Thus, wholesale price at the monopoly level is

consequently self-sustained.

Intuitively, cooperation to play the uniform retail policy eliminates net payments, which are

the source of deviation incentives. A similar cooperation in the context of balancing net

payments is found in Shy (2001) model on international settlement rates.

The game can be depicted in the following extensive form. The bold inner segments

represent the one-shot and repeated game equilibria. In the one-shot game, the retailer non-

cooperatively plays discriminatory retail policy, , and the wholesaler plays unilateral price

. If the game is repeated infinitely, uniform retail policy, , with wholesale monopoly

price fixing, , are the equilibria.

8Figure (C) Networks choice on retail policy in extensive form.

Wholesaler

Retailer

185

Appendix (D). Instrumental Variables

In this appendix, we evaluate the use of the Omani networks (Nawras and Oman-Mobile),

separately and jointly, as instrumental variables (IVs) in the Two-Stage Least Square (2SLS)

estimation of Chapter (6), for the following model.

( ) ( )

( )

(D)

The variables are defined in Table (6.1). The observations are limited to highly visited

markets as explained in Chapter (6).

In Table (D.1) below, model (D5) is the same 1SLS used in Chapter (7); and model (D6) is

the same 2SLS used in Chapter (6). In the first-stage regression, the coefficients of the

Omani networks IVs, separately and jointly, are positive and significant at 1% level. As

shown in Table (D.2), the associated coefficient has F-statistics above 10.

However, their implications on the endogenous regressor (i.e. wholesale prices charged to

Etisalat) in the 2SLS differ considerably. Only with Nawras IV we find a negative and

significant price coefficient. In contrast, the price coefficient is negative and insignificant with

Oman-Mobile IV, and incorrect (i.e. with positive sign) and insignificant when both IVs are

used.

We checked for the validity of using both Omani networks as IVs, using the over

identification restrictions test, where the null hypothesis states that all instruments are

uncorrelated with the error term, , of Eq. (D). As shown in Table (D.2), at 5% significance

level, we fail to reject the null that both IVs are valid.

We think the validity of both IVs comes from the ability of each IV to reflect cost differentials

across markets, as will be elaborated next.

Furthermore, we run Durbin and Wu-Hausman tests for the null hypothesis that the OLS

estimator is equivalent to the 2SLS estimator using the Omani networks IVs, separately and

186

jointly. If OLS and 2SLS are similar, then there is no need for the instrument because OLS is

efficient; in other words, the endogenous regressor, PRICEjt, should be treated as exogenous.

However, we know in our case that the OLS estimator is biased due to the endogeneity

problem. With these tests, we try to know which IV(s) should recommend using 2SLS. Only

with Nawras IV we find significance: we reject the null hypothesis at 1% significance level,

and conclude the OLS and the 2SLS estimators are different. Therefore, the use of 2SLS is

only recommended when Nawras IV is used.

Therefore, in Chapter (6), we proceeded in the 2SLS estimation with the price of Nawras as

an IV. Next, we provide an interpretation why Nawras can be a better IV compared to Oman-

Mobile.

A variable of a home network’s retail prices (other than Etisalat) is chosen as an IV based on

the assumption that the retail prices reflect actual costs of visited networks. If the IV has

discriminatory retail prices, as with Oman-Mobile, it should reflect actual costs within a

market and across markets. However, because a discriminatory retail price reflects the

embedded wholesale price, this IV does not overcome any bias at the wholesale level

stemming from preferential agreements (e.g. alliance discounts). This bias can be in the IV

itself (e.g. Oman-Mobile retail prices), in the endogenous regressor (e.g. wholesale prices

charged to Etisalat), or in both variables if preferential agreements differ between home

networks.

On the other hand, an IV using uniform retail pricing, as with Nawras, is assumed to reflect

the common costs of visited networks in a given market, providing cost differentials across

markets. Although it does not tell about costs asymmetry within a market, an IV using

uniform retail pricing is neutral to preferential agreements and hence it overcomes the bias

caused by preferential wholesale prices.

187

28Table (D.1) Instrumental variables for simple logit demand models with Oman-Mobile.

Market High Visited Markets

Instrumental Variable(s) Oman-Mobile & Nawras Oman-Mobile Nawras

Estimator 1SLS 2SLS 1SLS 2SLS 1SLS 2SLS

Dependent variable PRICEjt ( ) ( ) PRICEjt ( ) ( ) PRICEjt ( ) ( )

Model (D1) (D2) (D3) (D4) (D5) (D6)

CONSTANT 2.107*** -4.860*** 3.255*** -4.764*** 2.733*** -3.865***

(0.306) (0.253) (0.283) (0.253) (0.314) (0.330)

YEAR

-0.310** -0.031 -0.262* -0.071 -0.251* -0.060

(0.141) (0.103) (0.139) (0.100) (0.148) (0.128)


0.773** -0.797*** 0.518 -0.533** 0.655* -0.639**

(0.355) (0.258) (0.348) (0.250) (0.369) (0.318)

PRICEjt (network’s own price)

- 0.021 - -0.015 - -0.203***

- (0.027) - (0.027) - (0.037)

OMAN_MOBILEj (Oman-Mobile retail price in AED in 2013 for roaming on j)

0.150*** - 0.298*** - - -

(0.018) - (0.011) - - -


0.207*** - - - 0.262*** -

(0.018) - - - (0.011) -



-0.795 -0.024 -0.798 0.052 -1.050* -0.342

(0.605) (0.440) (0.596) (0.429) (0.635) (0.549)


0.459 0.135 0.476* 0.319 0.671** 0.439

(0.289) (0.211) (0.288) (0.207) (0.315) (0.273)


0.331 0.649*** -0.050 0.766*** 0.575** 1.042***

(0.216) (0.156) (0.208) (0.149) (0.225) (0.193)


CROSSOWNEDjt

1.341* -1.688*** 0.814 -1.621*** 1.513** -1.103*

(0.723) (0.526) (0.721) (0.518) (0.760) (0.655)

ALLIANCEj

-1.133*** 0.434 -1.081*** 0.392 -1.124** 0.133

(0.411) (0.300) (0.409) (0.295) (0.448) (0.388)

CAMELj -0.454 0.717*** -0.083 0.539*** -0.489 0.563**

(0.292) (0.212) (0.286) (0.205) (0.307) (0.264)

Observations 765 765 907 907 831 831

R2 0.509 - 0.442 - 0.413 -

R2 Adjusted 0.502 - 0.436 - 0.406 -


188

29Table (D.2) Tests for the instrumental variables used in Table (D.1).

Instrumental Variable(s) Oman-Mobile &

Nawras Oman-Mobile Nawras

Test/Result Score

or F-statistics

P-value Score

or F-statistics

P-value Score

or F-statistics

P-value

1SLS coefficient’s significance (H0: IV’s coefficient is insignificant)

374.195 0.000*** 689.083 0.000*** 550.561 0.000***

Over identification restrictions (H0: Both IVs are valid)

Sargan 5.105 0.024** - - - -

Basmann 5.066 0.024** - - - -

Price endogeneity: (H0: OLS and 2SLS are the same)

Durbin 1.703 0.192 0.748 0.388 37.980 0.000***

Wu-Hausman 1.682 0.195 0.739 0.390 39.272 0.000***

* p<0.10, ** p<0.05, *** p<0.01

Furthermore, we try Oman-Mobile IV, but after taking its average retail price in the relevant

market. By doing so, we try to eliminate what we call the wholesale bias, as explained

previously, by turning the discriminatory nature of the IV to something like uniform IV.

In Table (D.3) below, the number of observations for Oman-Mobile prices rises from 907 to

987, because the missing visited networks in Oman-Mobile IV take the market average. The

average price has a positive and significant coefficient at 1% level in the 1SLS models (D7)

and (D9). Notice the significance of the endogenous regressor, PRICEjt, in Table (D.3),

compared to the previous models (D2) and (D4). The sign of the coefficient is negative with

1% significance level. The size of the coefficients in models (D6), (D8) and (D10) is similar.

Moreover, model (D10) has very similar results to model (D6) regarding the rest of

regressors; and the same is true with model (D8) except for POLICY and ALLIANCEj.

Table (D.4) below presents test results related to the market average of Oman-Mobile.

Interestingly, the null hypothesis related to price endogeneity is rejected with 1% significance

level, compared to Table (D.2) when we failed to reject it. The lesson learned here is that the

discriminatory IV can be a good instrument if we take its market average, intuitively, because

of removing preferential wholesale bias.

189

30Table (D.3) Instrumental variables for simple logit demand models with average Oman-Mobile.

Market High Visited Markets

Instrumental Variable(s) Average Oman-Mobile Average Oman-Mobile &

Nawras

Estimator 1SLS 2SLS 1SLS 2SLS

Dependent variable PRICEjt ( ) ( ) PRICEjt ( ) ( )

Model (D7) (D8) (D9) (D10)

CONSTANT 2.754*** -3.954*** 2.498*** -3.976***

(0.288) (0.312) (0.313) (0.325)

YEAR

-0.212 -0.045 -0.252* -0.055 (0.136) (0.120) (0.146) (0.127)


0.455 -0.481 0.604* -0.647** (0.340) (0.299) (0.363) (0.316)

PRICEjt (network’s own price)

- -0.216*** - -0.186*** - (0.035) - (0.036)

AVERAGE_OMAN_MOBILEr (Oman-Mobile average retail price in AED in 2013 for roaming in market r)

0.315*** - 0.135*** -

(0.012) - (0.026) -


- - 0.172*** -

- - (0.021) -



-1.143* -0.272 -1.063* -0.315

(0.590) (0.521) (0.625) (0.545)


0.843*** 0.568** 0.754** 0.423 (0.294) (0.260) (0.310) (0.271)


0.259 1.086*** 0.437* 1.038*** (0.205) (0.180) (0.223) (0.192)


CROSSOWNEDjt

1.246* -1.101* 1.494** -1.124* (0.714) (0.629) (0.748) (0.651)

ALLIANCEj

-1.121*** 0.119 -1.127** 0.152 (0.418) (0.369) (0.441) (0.385)

CAMELj -0.152 0.417* -0.426 0.567**

(0.281) (0.247) (0.303) (0.262)

Observations 987 987 831 831

R2 0.407 - 0.431 -

R2 Adjusted 0.401 - 0.424 -


190

31Table (D.4) Tests for the instrumental variables used in Table (D.3).

Instrumental Variable(s) Average Oman-Mobile Average Oman-Mobile &

Nawras

Test/Result Score

or F-statistics

P-value Score

or F-statistics

P-value

1SLS coefficient’s significance (H0: IV’s coefficient is insignificant)

646.622 0.000*** 296.585 0.000***

Over identification restrictions (H0: Both IVs are valid)

Sargan - - 5.400 0.020**

Basmann - - 5.364 0.021**

Price endogeneity: (H0: OLS and 2SLS are the same)

Durbin 39.9823 0.000*** 32.8469 0.000***

Wu-Hausman 41.2059 0.000*** 33.7459 0.000***

* p<0.10, ** p<0.05, *** p<0.01

191

Appendix (E). Additional Estimations For Demand And Steering Models

This appendix provides additional estimations that are referred to in the demand models

(Chapter 6) and the effectiveness of steering models (Chapter 8). Those two chapters share

similarities in the fact that they have visited networks’ market shares as the dependent

variable.

The main purpose of this appendix is to support the argument that we should have a

demand model that is separate from a steering model such that the demand model reflects

demand-side substitution made by Etisalat roamers, while the steering model reflects

supply-side substitution made by the home network, Etisalat. In addition, the wholesale price

set by a visited network as given in the dataset (hereafter given price) suits the demand

model; and the wholesale price set by a visited network relative to its market arithmetic

mean (hereafter relative price) suits the steering model.

For the sake of the above argument, we will use the simple logit demand model similar to

Eq. (6.1) of Chapter (6) under different specifications using either the given price or the

relative price, with and without the interaction term of price and policy. Moreover, we will

compare estimations between highly visited markets, low visited markets and all markets.

Since we shall be testing for steering also and for the sake of the above argument, we found

the results are very similar if, instead, we used the functional form as in Chapter (8) for

steering models.

The model for the simple logit demand is provided by Eq. (E.1) with the given price, which is

estimated by Ordinary Least Squares (OLS). Subscript for year is dropped as data is pooled,

and market subscript is dropped for convenience.

( ) ( )

( )

(E.1)

192

( ) is the natural log of the market share of visited network ; ( ) is the natural log of

the outside good share; and is the error term. The product refers to the outside good

(see Table 6.2). All the variables that will be used in this appendix are listed in Table (E.1)

below.

In addition, we will also re-estimate the demand using the relative price in the following

model.

( ) ( )

( )

(E.2)

The key variable that can support the aforementioned argument lies in the interaction term of

the price and the policy dummy variable (1 if uniform; 0 if discriminatory), which tells us

about the effect of wholesale price on a visited network’s market share before and after

Etisalat retail pricing policy. The reason behind this is the consumer’s discretion regarding

price as assumed throughout the thesis. That is, before the policy, roamers care about price

(since the retail price reflects the wholesale price) and thus should choose manually

between visited networks; while after the policy, roamers leave handsets on automatic mode

since retail price is common to use any network in a given market (assuming roamers’

decisions on network selection are mainly based on price). Automatic network selection

allows Etisalat to choose between visited networks via its steering technologies. In other

words, before the policy, demand-side substitution is assumed; and after the policy, supply-

side substitution is assumed. Accordingly, the testable hypotheses are listed and explained

as follows.

H1:

o That is, before the policy (i.e. during the years when Etisalat retail price was

discriminatory by visited networks), the price coefficient is insignificant.

o If the coefficient is insignificant, we fail to reject H1 and conclude that roamers do not

care about price before the policy such that they leave their handsets on default-

automatic mode. This conclusion can also mean that roamers are price insensitive. In

addition, it means steering by Etisalat can be enabled, but is ineffective because

market share does not respond to changes in wholesale price.

193

o If the coefficient is negative and significant, we reject H1 and conclude that roamers

care about price before the policy such that they use their handsets to manually

select the visited network with the cheapest price. This also means that steering by

Etisalat is disabled because steering requires handsets to be on automatic mode.

H2: .

o That is, after the policy (i.e. during the years when Etisalat retail price became

uniform), the joint coefficients of price and the interaction term are insignificant.

o If the joint coefficients are insignificant, we fail to reject H2 and conclude that roamers

do not care about price after the policy such that they leave their handsets on default-

automatic mode. This conclusion also means steering by Etisalat can be enabled, but

is ineffective because market share does not respond to changes in wholesale price.

o If the joint coefficients are negative and significant, we reject H2 and conclude that

market share is responsive to wholesale price after the policy. Given automatic

network selection that enables steering, this conclusion is interpreted in terms of

buyer power as steering by Etisalat is effective because market share responds to

changes in wholesale price.

The results of testing the above hypotheses together will be interpreted to judge which

model should be considered a demand model, a steering model, and a demand or steering

model, or none of these (see Table E.0 below). Then based on such judgement, we justify

why we should use the given price and why we should use the relative price.

32Table (E.0) Possible scenarios for the hypotheses test results and our intuitions.

Possible hypotheses test results

Explanation Suggested use of the

model

Reject both H1 and H2 Price is significant before and after the policy

Demand or steering

Reject H1 but fail to reject H2

Price is significant only before the policy Demand only

Fail to reject H1 but reject H2

Price is significant only after the policy Steering only

Fail to reject both H1 and H2

Price is insignificant before and after the policy

None

Demand estimations are presented for highly visited markets in Table (E.2), for low visited

markets in Table (E.3), and for all markets (i.e. all observations) in Table (E.4). The marginal

effects of price and policy are reported in Table (E.5).

194

33Table (E.1) Notations and variables related to estimations.







POLICY Dummy variable equals 1 for the years 2010-2011 (when Etisalat implemented its uniform retail policy); 0 for the years 2008-2009 (when Etisalat implemented discriminatory retail policy).



RPRICEjrt j’s average wholesale price per minute of outgoing call relative to the arithmetic mean of its market r in year t (in constant AED with 2007 as base year).


Description





Description


Outside good market share in year t.

195

34Table (E.2) Simple logit demand models for observations in Highly Visited Markets. Dependent variable:

( ) ( )

(E1)

(E2)

(E3)

(E4)

(E5)

(E6)

(E7)

(E8)

CONSTANT

-4.305*** -5.112*** -5.006*** -4.749*** -4.352*** -5.320*** -5.283*** -5.743***

(0.253) (0.245) (0.252) (0.307) (0.289) (0.279) (0.284) (0.466)

YEAR

-0.028 -0.006 -0.007 -0.006 -0.013 0.008 0.006 0.004

(0.125) (0.116) (0.116) (0.116) (0.126) (0.117) (0.116) (0.116)


-0.379 -0.328 -0.526* -0.916** -0.392 -0.342 -0.539* 0.052

(0.282) (0.262) (0.290) (0.393) (0.283) (0.262) (0.290) (0.556)

PRICEjt (network own price)

-0.056** -0.055*** -0.052** -0.095*** - - - -

(0.023) (0.021) (0.021) (0.036) - - - -

RPRICEjrt (network own price relative to arithmetic mean)

- - - - -0.313 -0.147 -0.060 0.411

- - - - (0.203) (0.189) (0.192) (0.424)



- -0.912*** -0.003 -0.070 - -0.847*** 0.069 0.140

- (0.285) (0.504) (0.506) - (0.286) (0.506) (0.509)


- 0.621*** 0.465* 0.488** - 0.610*** 0.440* 0.425*

- (0.177) (0.248) (0.249) - (0.178) (0.249) (0.249)


- 1.328*** 1.092*** 1.088*** - 1.330*** 1.098*** 1.086***

- (0.119) (0.174) (0.174) - (0.120) (0.175) (0.175)


CROSSOWNEDjt

- - -1.286** -1.208** - - -1.318** -1.357**

- - (0.610) (0.612) - - (0.617) (0.617)

ALLIANCEj

- - 0.293 0.281 - - 0.335 0.330

- - (0.353) (0.353) - - (0.356) (0.355)

CAMELj

- - 0.407* 0.412* - - 0.400* 0.408*

- - (0.240) (0.239) - - (0.241) (0.241)

PRICEjt

- - - 0.065 - - - -

- - - (0.044) - - - -

RPRICEjrt

- - - - - - - -0.592

- - - - - - - (0.475)

Observations 995 995 995 995 995 995 995 995

R2 0.017 0.158 0.164 0.166 0.013 0.152 0.159 0.161

R2 Adjusted 0.0141 0.153 0.157 0.158 0.0105 0.147 0.152 0.152


196

35Table (E.3) Simple logit demand models for observations in Low Visited Markets. Dependent variable:

( ) ( )

(E9)

(E10)

(E11)

(E12)

(E13)

(E14)

(E15)

(E16)

CONSTANT

-3.856*** -3.760*** -3.754*** -3.715*** -2.621*** -2.512*** -2.583*** -3.613***

(0.201) (0.199) (0.199) (0.233) (0.241) (0.240) (0.240) (0.404)

YEAR

-0.102 -0.101 -0.100 -0.100 -0.115 -0.108 -0.107 -0.107

(0.095) (0.094) (0.093) (0.093) (0.094) (0.092) (0.091) (0.091)


-0.310 -0.285 -0.319 -0.386 -0.303 -0.281 -0.306 1.037**

(0.212) (0.209) (0.212) (0.299) (0.210) (0.205) (0.208) (0.473)


0.038** 0.024 0.026 0.020 - - - -

(0.016) (0.016) (0.016) (0.024) - - - -


- - - - -0.955*** -1.055*** -0.972*** 0.047

- - - - (0.180) (0.178) (0.178) (0.368)



- -1.314*** 0.133 0.120 - -1.501*** -0.207 0.086

- (0.236) (0.400) (0.402) - (0.231) (0.396) (0.405)


- -0.230 -0.439 -0.440 - -0.295 -0.373 -0.448

- (0.226) (0.326) (0.326) - (0.222) (0.322) (0.321)


- 0.281** -0.031 -0.028 - 0.236* -0.046 -0.015

- (0.129) (0.182) (0.183) - (0.127) (0.179) (0.179)


CROSSOWNEDjt

- - -2.160*** -2.139*** - - -1.921*** -2.231***

- - (0.489) (0.494) - - (0.484) (0.492)

ALLIANCEj

- - 0.387 0.393 - - 0.147 0.175

- - (0.446) (0.447) - - (0.442) (0.440)

CAMELj

- - 0.575** 0.569** - - 0.533** 0.479*

- - (0.255) (0.256) - - (0.252) (0.251)

PRICEjt

- - - 0.010 - - - -

- - - (0.032) - - - -

RPRICEjrt

- - - - - - - -1.326***

- - - - - - - (0.420)

Observations 932 932 932 932 932 932 932 932

R2 0.039 0.074 0.099 0.099 0.062 0.106 0.125 0.134

R2 Adjusted 0.0360 0.0681 0.0904 0.0895 0.0587 0.100 0.116 0.125


197

36Table (E.4) Simple logit demand models for all observations. Dependent variable:

( ) ( )

(E17)

(E18)

(E19)

(E20)

(E21)

(E22)

(E23)

(E24)

CONSTANT

-4.266*** -4.452*** -4.397*** -4.339*** -3.618*** -3.876*** -3.905*** -4.710***

(0.168) (0.170) (0.172) (0.205) (0.197) (0.198) (0.200) (0.334)

YEAR

-0.043 -0.035 -0.037 -0.037 -0.045 -0.036 -0.039 -0.041

(0.082) (0.081) (0.080) (0.080) (0.082) (0.080) (0.080) (0.080)


-0.349* -0.310* -0.410** -0.505* -0.348* -0.310* -0.401** 0.635

(0.184) (0.181) (0.190) (0.263) (0.183) (0.180) (0.189) (0.394)


0.012 0.010 0.012 0.003 - - - -

(0.014) (0.014) (0.014) (0.022) - - - -


- - - - -0.566*** -0.505*** -0.411*** 0.398

- - - - (0.142) (0.140) (0.141) (0.304)



- -1.011*** 0.344 0.328 - -1.018*** 0.235 0.408

- (0.200) (0.346) (0.348) - (0.198) (0.346) (0.350)


- 0.291** 0.101 0.103 - 0.261* 0.122 0.086

- (0.145) (0.205) (0.205) - (0.145) (0.204) (0.204)


- 0.521*** 0.278** 0.277** - 0.508*** 0.278** 0.275**

- (0.088) (0.126) (0.126) - (0.087) (0.126) (0.126)


CROSSOWNEDjt

- - -1.984*** -1.961*** - - -1.840*** -1.987***

- - (0.422) (0.424) - - (0.423) (0.425)

ALLIANCEj

- - 0.359 0.361 - - 0.269 0.271

- - (0.288) (0.288) - - (0.289) (0.288)

CAMELj

- - 0.422** 0.424** - - 0.407** 0.402**

- - (0.175) (0.175) - - (0.174) (0.174)

PRICEjt

- - - 0.015 - - - -

- - - (0.028) - - - -

RPRICEjrt

- - - - - - - -1.029***

- - - - - - - (0.343)

Observations 1,927 1,927 1,927 1,927 1,927 1,927 1,927 1,927

R2 0.015 0.050 0.064 0.064 0.023 0.056 0.068 0.072

R2 Adjusted 0.0137 0.0468 0.0598 0.0595 0.0213 0.0529 0.0636 0.0675


198

37Table (E.5) Marginal effects on visited networks’ prices post Etisalat uniform retail policy.

Observations Null

Hypothesis Relevant model F-statistics P-value

Observations in High Visited Markets (Table E.2)

Given price (E4) 1.35 0.245

Relative price (E8) 0.71 0.400

Observations in Low Visited Markets (Table E.3)

Given price (E12) 2.04 0.154

Relative price (E16) 40.08 0.000***

All observations (Table E.4)

Given price (E20) 0.98 0.321

Relative price (E24) 15.79 0.000***

* p<0.10, ** p<0.05, *** p<0.01

We discuss models (E4) and (E8) for the highly visited markets (Table E.2), models (E12)

and (E16) for the low visited markets (Table E.3), and models (E20) and (E24) for all

observations (Table E.4). The marginal effects post-the-policy are discussed based on Table

(E.5). Conclusions regarding which model and price suits to model demand or steering are

drawn in light of Table (E.0).

For visited networks in the highly visited markets, the given price is found negative and

significant only before the policy (model E4); while the relative price is found insignificant

before or after the policy (model E8). Therefore, we can reject H1 but fail to reject H2 with

the given price; however with the relative price, we fail to reject both H1 and H2.

We conclude that the model for highly visited markets with the given price, model (E4),

should be used to model demand, as roamers care about the wholesale price only during the

time when it was reflected in the retail price (i.e. before the policy when retail prices were

discriminatory). However, no model for highly visited markets suits to model steering, as the

market share does not respond to changes in wholesale price after the policy. Moreover, the

relative price does not suit the demand model as it is insignificant before or after the policy,

which can be due to the fact that relative price does not reflect the level of prices to which

roamers are sensitive.

For visited networks in the low visited markets, the given price is found insignificant before or

after the policy (model E12); while the relative price is found negative and significant only

after the policy (E16). Therefore, we fail to reject H1 and H2 with the given price; however

with the relative price, we fail to reject H1 but we can reject H2.

199

We conclude that the model for low visited markets with the relative price, model (E16),

should be used to model steering. This is so because market share is responsive to changes

in wholesale price, suggesting that Etisalat has effective steering after its uniform retail policy

that makes undercutting by a visited network attractive to raise own market share. However,

no model for low visited markets suits to model demand, as the market share does not

respond to changes in wholesale price before the policy. Moreover, the given price does not

suit the steering model as it is insignificant before or after the policy, because probably

roamers are insensitive to price in the low visited markets. In addition, the same intuitions for

the low visited markets are true for all observations (models E20 and E24).

In sum, we think the given price suits the demand model as it represents roamers’ response

to price (i.e. demand-side substitution), while relative price suits the steering model as it

represents Etisalat’s response to price (i.e. supply-side substitution).

A part from the hypotheses, the coefficient of the given price in the low visited markets is

found positive and significant (model E9). This is caused by not controlling for visited

network’s characteristics. In fact, CAMEL dummy variable’s coefficient is only significant

after the policy. However, the significance and size of the CAMEL dummy variable’s

coefficient indicate that low visited markets do not significantly host prepaid roamers as

compared to the highly visited markets.

200

Appendix (F). Mean Elasticities In EEA Countries 38Table (F) Estimated mean own price elasticity of demand for EEA visited networks by market.

Market Mean elasticity

(2008-2011)

AUSTRIA -1.428

BELGIUM -1.659

CYPRUS -0.511

CZECH REPUBLIC -2.018

DENMARK -1.174

FINLAND -0.931

FRANCE -1.640

GERMANY -1.244

GREECE -1.127

IRELAND -1.140

ITALY -1.490

NETHERLANDS -1.176

NORWAY -0.718

PORTUGAL -1.548

ROMANIA -1.689

SPAIN -1.458

SWEDEN -1.020

UK -1.178

All EEA -1.295 Source: Estimated elasticity based on Eq. (6.2) using the price coefficient of model (6f).

201

Appendix (G). List Of Visited Networks 39Table (G) List of all foreign networks visited by Etisalat roamers during the years 2008-2011.

SERIAL TAP OPERATOR MARKET SERIAL TAP OPERATOR MARKET

1 AFGAW AFGHAN WIRELESS AFGHANISTAN 51 BIHER HT ERONET BOSNIA HERZEGOVINA

2 AFGEA ETISALAT-AFGHN AFGHANISTAN 52 BIHMS MOBILNA - BIH BOSNIA HERZEGOVINA

3 AFGAR MTN AFGHANISTAN 53 BWAGA MASCOM-BOTSWANA BOTSWANA

4 AFGTD TDC(ROSHAN) AFG AFGHANISTAN 54 BWAVC VISTA CELLULAR BOTSWANA

5 ALBAM AMC, ALBANIA ALBANIA 55 BRABT 14 BRASIL TELCO BRAZIL

6 ALBEM EAGLE MOBILE ALBANIA 56 BRACS TIM - BRACS BRAZIL

7 ALBVF VODAFONE ALBANI ALBANIA 57 BRARN TIM - BRARN BRAZIL

8 DZAA1 ATM MOBILIS ALG ALGERIA 58 BRASP TIM - BRASP BRAZIL

9 DZAOT DJEZZY ALGERIA 59 BRATM TNL PCS BRAZIL BRAZIL

10 DZAWT WATANIYA - DZA ALGERIA 60 BRATC VIVO-MG BRAZIL

11 ANDMA MOBILAND GSM ANDORRA 61 BRNBR B.MOBILE COMM. BRUNEI

12 AGOUT UNITEL ANGOLA 62 BRNDS DST COMMUNICATI BRUNEI

13 AIACW C&W ANGUILLA ANGUILLA 63 BGRVA BTC BULGARIA

14 ATGCW C&W ANTIGUA ANTIGUA & BARBUDA 64 BGRCM GLOBUL BULGARIA

15 ARGTP TELE. ARGENTINA ARGENTINA 65 BGR01 MOBILTEL,BULGRI BULGARIA

16 ARM01 ARMENTEL GSM900 ARMENIA 66 BFACT CELTEL - B.FASO BURKINA FASO

17 ARM05 K-TEL VIVACELL ARMENIA 67 BFATL TELECEL - BFASO BURKINA FASO

18 ARMOR ORANGE ARMENIA ARMENIA 68 BFAON TELMOB BURKINA FASO

19 AUSHU HUTCHISON 3G AU AUSTRALIA 69 BDIET ECONET WIRELESS BURUNDI

20 AUSOP OPTUS,AUSTRALIA AUSTRALIA 70 KHMCC CADCOMMS CAMBODIA

21 AUSTA TELSTRA, AUS AUSTRALIA 71 KHMGM CAMGSM CAMBODIA

22 AUSVF VODAFONE-AUS AUSTRALIA 72 KHMSM CASACOM CAMBODIA

23 AUTHU H3G AUSTRIA 73 KHML1 LATEZ CO, LTD CAMBODIA

24 AUTPT MOBILKOM, AUT AUSTRIA 74 KHMSH SHINAWTRA CO. CAMBODIA

25 AUTCA ORANGE AUSTRIA AUSTRIA 75 CMRMT MTN - CAMEROON CAMEROON

26 AUTMM T-MOBIL AUSTRIA AUSTRIA 76 CMR02 ORANGE CAMEROUN CAMEROON

27 AZEAC AZERCEL - AZB AZERBAIJAN 77 CANBM BELL MOBILITY CANADA

28 AZEAF AZERFON LLC AZERBAIJAN 78 CANRW ROGERS - CANADA CANADA

29 AZEBC BAKCELL LTD AZERBAIJAN 79 CANTS TELUS CANADA

30 BHRBT BATELCO,BAHRAIN BAHRAIN 80 CYMCW C&W CAYMAN CAYMAN

31 BHRST STC BAHRAIN BAHRAIN 81 CAFAT MOOV-RCA CENTRAL AFRICA REPUBLIC

32 BHRMV ZAIN-BAHRAIN BAHRAIN 82 TCDCT CELTEL - TCHAD CHAD

33 BGDGP GRAMEEN PHONE BANGLADESH 83 TCDML MILICOM (TIGO) CHAD

34 BGDBL SHEBA TELECOM BANGLADESH 84 CHLMV ENTEL PCS TELEC CHILE

35 BGDTT TELETALK BNGDSH BANGLADESH 85 CHNCT CHINA TELECOM CHINA

36 BGDAK TM INTERNATIONL BANGLADESH 86 CHNCU CHINA UNICOM CHINA

37 BGDWT WARID(BANGLADSH BANGLADESH 87 COLCM COM.COMCEL S.A COLOMBIA

38 BRBCW C&W BARBADOS BARBADOS 88 COMHR COMORES TELECOM COMOROS ISLANDS

39 BLRMD FE 'VELCOM' BELARUS 89 CODCT CELTEL-DRC CONG CONGO (DEMOCRATIC REPUBLIC)

40 BLR02 MTS - BELARUS BELARUS 90 CODSA OASIS SPRL CONGO (DEMOCRATIC REPUBLIC)

41 BELTB BELGACOM - BEL BELGIUM 91 CODVC VODACOM - CONGO CONGO (DEMOCRATIC REPUBLIC)

42 BELKO KPN(BASE) BELGIUM 92 COGCT CELTEL - CONGO CONGO (PEOPLES REPUBLIC)

43 BELMO MOBISTAR GSM BELGIUM 93 COGLB LIBERTIS - COG CONGO (PEOPLES REPUBLIC)

44 BENSP AREEBA - BENIN BENIN 94 COGWC WARID CONGO S.A CONGO (PEOPLES REPUBLIC)

45 BEN02 MOOV-BENIN BENIN 95 HRVCN T-MOBILE CROTIA CROATIA

46 BMUNI M3 WITELESS LTD BERMUDA 96 HRVVI VIP NET GSM CROATIA

47 BTNBM B-MOBILE(BTNBM) BHUTAN 97 CUB01 C_COM (CUBACEL) CUBA

48 BOLME MOVIL DE ENTEL BOLIVIA 98 CYPCT CYTA, CYPRUS CYPRUS

49 BOLNT NUEVATEL BOLIVIA 99 CYPSC MTN CYPRUS CYPRUS

50 BIHPT GSMBIH - BOSNIA BOSNIA HERZEGOVINA 100 CZECM OSKAR CZECH REPUBLIC

202

Table (G) List of all foreign networks visited by Etisalat roamers during the years 2008-2011.


101 CZERM T-MOBILE CZECH CZECH REPUBLIC 151 GRCSH TIM - GREECE GREECE

102 CZEET TELEFONICA O2 CZECH REPUBLIC 152 GRCPF VODAFONE PANAFO GREECE

103 DNKHU HI3G ACCESS-DNK DENMARK 153 GRDCW C&W GRENADA GRENADA

104 DNKDM SONOFON,DENMARK DENMARK 154 GUMHT GUAM WIELESS GUAM

105 DNKTD TDC MOBIL A/S DENMARK 155 GTMSC SERCOM (CLARO) GUATEMALA

106 DNKIA TELIA-DENMARK DENMARK 156 GBRGT GUERNSEY, UK GUERNSEY

107 DJIDJ DJIBOUTI-TELECO DJIBOUTI 157 GIN07 CELLCOM GUINEA GUINEA

108 DMACW C&W DOMINICA DOMINICA 158 GIN03 INTERCEL GUINEE GUINEA

109 DOM01 ORANGE DOMINICA DOMINICAN REPUBLIC 159 GINGS ORANGE GUINEE GUINEA

110 EGYEM ETISALAT MISR EGYPT 160 GUYUM CEL*STAR GUYANA GUYANA

111 EGYAR MOBINIL, EGYPT EGYPT 161 HKGPP CHINA MBLE PTCL HONGKONG

112 EGYMS VODAFONE EGYPT EGYPT 162 HKGNW CSL LTD HONGKONG

113 SLVTP STE TELECOM SLV EL SALVADOR 163 HKGTC CSL, HONG KONG HONGKONG

114 SLVTM TELEMOVIL EL SL EL SALVADOR 164 HKGMC HKT(PCCW MBL) HONGKONG

115 GNQ01 ORANGE GQ EQUATORIAL GUINEA 165 HKGM3 HKT(PCCW MBLE) HONGKONG

116 ESTEM AS EMT(EMT GSM) ESTONIA 166 HKGH3 HUTCHISON - HKG HONGKONG

117 ESTRE ELISA MOBIILSID ESTONIA 167 HKGHT HUTCHISON, HKG HONGKONG

118 ESTRB TELE2 - ESTONIA ESTONIA 168 HKGSM SMARTONE-VODAFN HONGKONG

119 ETH01 ETHIOPIAN TELE ETHIOPIA 169 HUNH2 MAGYAR TELEKOM HUNGARY

120 FROFT FAROESE TELE FAROE ISLANDS 170 HUNH1 PANNON GSM HUNGARY

121 FJIDP DIGICEL(FIJI) FIJI 171 HUNVR VODAFONE-HUN HUNGARY

122 FJIVF VF FIJI LIMITED FIJI 172 ISLPS SIMINN -ICELAND ICELAND

123 FIN2G DNA FINLAND FINLAND 173 ISLTL VODAFONE ICELAND

124 FINRL ELISA CORPORTN. FINLAND 174 INDJH BHARTI, ANDHRA INDIA, Andhra Pradesh

125 FINTF SONERA FINLAND FINLAND 175 IND23 DISHNET-AIRCELL INDIA, Andhra Pradesh

126 FINAM ALANDS MOBITEL FINLAND, ALANDS 176 INDAH ETISALAT DB INDIA, Andhra Pradesh

127 FRAF3 BOUYGUES GSM FRANCE 177 IND07 IDEA CELLULAR INDIA, Andhra Pradesh

128 FRAF1 FRANCE TELECOM FRANCE 178 INDT0 TATA TELESERVCE INDIA, Andhra Pradesh

129 FRAF2 SFR - CEGETEL FRANCE 179 IND04 DISHNET-AIRCELL INDIA, Bihar

130 FRAF4 BOUYGUES TELECM FRENCH WEST INDIES 180 INDBR ETISALAT DB INDIA, Bihar

131 GUF01 OUTREMER TELECO FRENCH WEST INDIES 181 INDIB IDEAR CELLULAR INDIA, Bihar

132 GABCT CELTEL - GABON GABON 182 INDTB TATA TELESERVIC INDIA, Bihar

133 GAB01 LIBERTIS-GABON GABON 183 INDRC AIRCEL CELLULAR INDIA, Chennai

134 GABTL MOOV-GABON GABON 184 INDSC SKYCELL COMM INDIA, Chennai

135 GMBAC AFRICELL-GAMBIA GAMBIA 185 IND19 AIRCELL LTD INDIA, Delhi

136 GMBCM COMIUM GAMBIA GAMBIA 186 INDAT AIRTEL, INDIA INDIA, Delhi

137 GMB01 GAMCEL - GAMBIA GAMBIA 187 INDE1 ESSAR MOBILE INDIA, Delhi

138 GEOGC GEOCEL,GEORGIA GEORGIA 188 INDND ETISALAT DB INDIA, Delhi

139 GEOMA MAGTICOM,GEO GEORGIA 189 INDID IDEA CELLULAR INDIA, Delhi

140 GEOMT MOBITEL GEORGIA GEORGIA 190 INDDL MTNL - DELHI INDIA, Delhi

141 DEUE1 E-PLUS, GERMANY GERMANY 191 INDA3 B.A.GUJARAT INDIA, Gujarat

142 DEUE2 O2 (GERMANY) GM GERMANY 192 INDGU EISALAT DB INDIA, Gujarat

143 DEUD1 T-MOBILE,GERMAN GERMANY 193 INDF1 ESSAR GUJARAT INDIA, Gujarat

144 DEUD2 VODAFONE GERMAN GERMANY 194 INDBI IDEA CELLULAR INDIA, Gujarat

145 GHAMT MILLICOM GHANA 195 INDTG TATA TELESERVIC INDIA, Gujarat

146 GHASC MTN (SCANCOM) GHANA 196 INDA5 B.A.HARYANA INDIA, Haryana

147 GHAGT ONETOUCH - GHA GHANA 197 INDHY ETISALAT DB INDIA, Haryana

148 GHAZN ZAIN COM.GHANA GHANA 198 INDEH IDEA CELLULAR INDIA, Haryana

149 GIBGT GIBTEL, GIB GIBRALTAR 199 INDTH TATA TELESERVIC INDIA, Haryana

150 GRCCO COSMOTE GSM GREECE 200 INDBL AIRTEL,HIMACHAL INDIA, Himachal Pradesh

203



201 IND02 DISHNET AIRCELL INDIA, Himachal Pradesh 251 INDTC ETISALAT DB INDIA, Tamilnadu

202 INDIH IDEA CELLULAR INDIA, Himachal Pradesh 252 INDIT IDEA CELLULAR INDIA, Tamilnadu

203 IND22 AIRCELL LIMITED INDIA, Karnataka 253 INDT2 TATA TELESERVIC INDIA, Tamilnadu

204 INDJB BHARTI,KARNATAK INDIA, Karnataka 254 IND17 DISHNET AIRCELL INDIA, Uttar Pradesh East

205 INDKA ETISALAT DB INDIA, Karnataka 255 INDPE ETISALAT DB INDIA, Uttar Pradesh East

206 INDSK SPICE TELECOM-K INDIA, Karnataka 256 INDIU IDEA CELLULAR INDIA, Uttar Pradesh East

207 INDT1 TATA TELESERVIC INDIA, Karnataka 257 INDT7 TATA TELESRVICE INDIA, Uttar Pradesh East

208 INDA7 BHARTI KERALA INDIA, Kerala 258 INDA6 B.A.UP INDIA, Uttar Pradesh West

209 IND20 DISHNET AIRCELL INDIA, Kerala 259 IND18 DISHNET AIRCELL INDIA, Uttar Pradesh West

210 INDBK ESSAR CELLULAR INDIA, Kerala 260 INDPW ETISALAT DB INDIA, Uttar Pradesh West

211 INDKE ETISALAT DB INDIA, Kerala 261 INDEU IDEA CELLULAR INDIA, Uttar Pradesh West

212 INDEK IDEA CELLULAR INDIA, Kerala 262 INDT8 TATA TELESERVCE INDIA, Uttar Pradesh West

213 INDT3 TATA TELESERVIC INDIA, Kerala 263 INDWB BSNL - INDIA INDIA, West Bengal

214 INDMT BHARTI MOBITEL INDIA, Kolkata 264 IND03 DISHNET AIRCELL INDIA, West Bengal

215 INDCC EAST LIMITED INDIA, Kolkata 265 INDIW IDEA CELLULAR INDIA, West Bengal

216 INDIK IDEA CELLULAR INDIA, Kolkata 266 INDT9 TATA TELESERVIC INDIA, West Bengal

217 INDTK TATA TELESERVIC INDIA, Kolkata 267 IDNLT AXIS(LIPPOTEL) INDONESIA

218 INDA8 B.A.MP INDIA, Madhya Pradesh 268 IDNEX EXCELCOMINDO INDONESIA

219 INDMD ETISALAT DB INDIA, Madhya Pradesh 269 IDN89 HUTCHISON CP INDONESIA

220 INDMP IDEA CELLULAR INDIA, Madhya Pradesh 270 IDNSL SATELINDO INDONESIA

221 INDRM RELIANCE INDIA INDIA, Madhya Pradesh 271 IDNTS TELKOMS, IDA INDONESIA

222 INDT5 TATA TELESERVIC INDIA, Madhya Pradesh 272 IRNMI MTN IRANCELL IRAN

223 IND24 AIRCELL LTD INDIA, Maharashtra 273 IRNRI TALIYA IRAN IRAN

224 INDA2 B.A.MAHARASHTA INDIA, Maharashtra 274 IRN11 TCI-GSM 900 IRAN

225 INDBM ESSAR CELLULAR INDIA, Maharashtra 275 IRNKI TKC-KIFZO, IRAN IRAN, KISH ISLAND

226 INDMA ETISALAT DB INDIA, Maharashtra 276 IRQAC ASIA CELL-IRAQ IRAQ

227 INDBO IDEA CELLULAR INDIA, Maharashtra 277 IRQOR IRAQNA (ZAIN) IRAQ

228 INDT6 TATA TELESERVIC INDIA, Maharashtra 278 IRQKK KOREK TELECOM IRAQ

229 IND21 AIRCELL LTD INDIA, Mumbai 279 IRQAT ZAIN-IRAQ IRAQ

230 INDA1 B.A.MUMBAI INDIA, Mumbai 280 IRLH3 HUTCHISON 3G IRELAND

231 INDB1 BPL MOBILE-MUM INDIA, Mumbai 281 IRLME METEOR, IRELAND IRELAND

232 INDHM ESSAR LIMITED INDIA, Mumbai 282 IRLDF O2 - IRELAND IRELAND

233 INDMU ETISALAT DB INDIA, Mumbai 283 IRLEC VODAFONE-IRELND IRELAND

234 INDIM IDEA CELLULAR INDIA, Mumbai 284 GBRMT MANX TELECOM,UK ISLE OF MAN

235 INDMB MTNL - MUMBAI INDIA, Mumbai 285 ITAH3 H3G S.P.A ITALY ITALY

236 INDTM TATA TELESERVIC INDIA, Mumbai 286 ITASI TIM - ITALIA ITALY

237 IND05 DISHNET AIRCELL INDIA, Orissa 287 ITAOM VODAFON-OMNITEL ITALY

238 INDIO IDEA CELLULAR INDIA, Orissa 288 ITAWI WIND - ITALY ITALY

239 INDTO TATA TELESERVIC INDIA, Orissa 289 CIVTL LOTENY TELECOM IVORY COST

240 INDPN ETISALAT DB INDIA, Punjab 290 CIV02 MOOV(A-CELL) IVORY COST

241 INDA9 PUNJAB - BHARTI INDIA, Punjab 291 CIV03 ORANGE-IVORY C IVORY COST

242 INDSP SPICE TELECOM-P INDIA, Punjab 292 JAMCW CABLE&WIRELESS JAMAICA

243 INDTP TATA TELESERVIC INDIA, Punjab 293 JAMDC DIGICEL-JAMAICA JAMAICA

244 INDRS ETISALAT DB INDIA, Rajasthan 294 JPNDO NTT DOCOMO-JPN JAPAN

245 INDH1 HEXACOM INDIA INDIA, Rajasthan 295 JPNJP SOFTBANK MOBLE JAPAN

246 INDIR IDEA CELLULAR INDIA, Rajasthan 296 GBRAJ AIRTEL JERSEY

247 INDTR TATA TELESERVIC INDIA, Rajasthan 297 GBRJT JERSY TELECOM JERSEY

248 INDAC AIRCELL LTD. INDIA, Tamilnadu 298 JORMC ORANGE JORDAN

249 INDA4 B.A.TAMIL NADU INDIA, Tamilnadu 299 JORUM UMNIAH JORDAN JORDAN

250 INDBT ESSAR CELLULAR INDIA, Tamilnadu 300 JORXT XPRESS - JORDAN JORDAN

204



301 JORFL ZAIN,JORDAN JORDAN 351 MLTGO GO MOBILE MALTA

302 ARMKT KARABAKH - ARM KARABAKH (DISPUTED REGION) 352 MLTMM MELITA MOBILE MALTA

303 KAZKZ K.CELL KAZAKHSTAN 353 MLTTL VDF MALTA LTD MALTA

304 KAZKT KAR-TEL LLP(KM) KAZAKHSTAN 354 MRTMT MATTEL MAURITANIA

305 KAZ77 MOBILE TELECOM KAZAKHSTAN 355 MRTMM MAURITEL MOBILE MAURITANIA

306 KENEC ESSAR TELECOM KENYA 356 MUSCP CELLPLUS,MTS MAURITIUS

307 KENSA SAFARICOM KENYA 357 MUSEM EMTEL LIMITED MAURITIUS

308 KENTK TELKOM KENYA KENYA 358 MEXTL RADIOMOVL(TELCL MEXICO

309 KENKC ZAIN-KENYA KENYA 359 MEXMS TELEFONICA MVLS MEXICO

310 KWTKT KUWAIT_KTC KUWAIT 360 MDAMC MOLDCELL MOLDOVA

311 KWTNM WATANIYA TELE KUWAIT 361 MDAVX VOXTEL MOLDOVA

312 KWTMT ZAIN,KUWAIT KUWAIT 362 MCOM1 MONACO TELECOM MONACO

313 KGZ01 BITEL,KYRGYZSTN KYRGYZSTAN 363 MNGMC MOBICOM-MONGOLI MONGOLIA

314 KGZNT NURTELECOM LLC KYRGYZSTAN 364 MNEMT MTEL MONTENEGRO

315 LAOTL TANGO LAO 365 YUGPM PRO MONTE GSM MONTENEGRO

316 LVABT BITE LATVIA LATVIA 366 YUGTM T-MOBILE MONTENEGRO

317 LVALM LATVIA MOBILE LATVIA 367 MSRCW MONTSERRAT MONTSERRAT

318 LVABC TELE2 - LATVIA LATVIA 368 MARMT MEDI TELECOM SA MOROCCO

319 LBNFL MIC 1 (ALFA) LEBANON 369 MARM1 MOROCO TELECOM MOROCCO

320 LBNLC MTC - LEBANON LEBANON 370 MARM3 WANA CORPORATE MOROCCO

321 LBR07 CELLCOM TELCO. LIBERIA 371 MOZ01 MCEL (TDM) MOZAMBIQUE

322 LBRCM COMIUM LIBERIA LIBERIA 372 MOZVC VODACOM - MOZ MOZAMBIQUE

323 LBY01 ELMADAR ALJADID LIBYA 373 NAM03 CELL ONE NAMIBIA

324 LBYLM LIBYANA MOBPHON LIBYA 374 NAM01 NAMIBIA TELECOM NAMIBIA

325 LIEMK MOBILKOM GSM LIECHTENSTEIN 375 NPLM2 SPICE NEPAL NEPAL

326 LIEVE ORANGE-LIEVE LIECHTENSTEIN 376 NLDPT KPN TELECOM NETHERLANDS

327 LIETG TELE 2 AKTIENGE LIECHTENSTEIN 377 NLDDT T-MBLE(ORANGE) NETHERLANDS

328 LTUOM OMNITEL, LIT LITHUANIA 378 NLDPN T-MOBILE - NLD NETHERLANDS

329 LTU03 TELE2-LITHUANIA LITHUANIA 379 NLDLT VODAFONE-NLD NETHERLANDS

330 LTUMT UAB BITE LITHUANIA 380 NZLNH NZ-COMMUNICATN NEW ZEALAND

331 LUXPT P&T, LUXEMBOURG LUXEMBOURG 381 NZLTM TELECOM MOBILE NEW ZEALAND

332 LUXTG TANGO-LUXEMBURG LUXEMBOURG 382 NZLBS VODAFON N.ZELND NEW ZEALAND

333 LUXVM VOX.MOBILE- LUX LUXEMBOURG 383 NICEN ENITEL-GSM1900 NICARAGUA

334 MACCT CTM, MACAU MACAU 384 NERCT CELTEL - NIGER NIGER

335 MACHT HUTCHISON-MACAU MACAU 385 NERTL MOOV-NIGER NIGER

336 MKDCC COSMOFON - MKD MACEDONIA 386 NGAEM ETISALT NIGERIA NIGERIA

337 MKDMM T-MBLE MACEDNIA MACEDONIA 387 NGAGM GLO MOBILE -NGA NIGERIA

338 MKDNO VIP OPERATOR MACEDONIA 388 NGAMN MTN NIGERIA NIGERIA

339 MDGCO CELTEL-MADACOM MADAGASCAR 389 NGAET ZAIN-NIGERIA NIGERIA

340 MDGAN ORANGE - MDGAN MADAGASCAR 390 NORNC NETCOM, NORWAY NORWAY

341 MDGTM TELMA MOBILE MADAGASCAR 391 NORNN NETWORK NORWAY NORWAY

342 MWICT CELTEL - MALAWI MALAWI 392 NORTM TELENOR, NORWAY NORWAY

343 MWICP TELEKOM MALAWI MALAWI 393 OMNNT NAWRAS - OMAN OMAN

344 MYSCC CELCOM, MYA MALAYSIA 394 OMNGT OMAN MOBILE OMAN

345 MYSMT DIGI TELECOM MALAYSIA 395 PAKMK MOBILINK, PAK PAKISTAN

346 MYSBC MAXIS, MALAYSIA MALAYSIA 396 PAKTP TELENOR - PAK PAKISTAN

347 MDV01 DHIRAAGU MALDIVES 397 PAKUF UFONE - PAK PAKISTAN

348 MDVWM WATANIYA-MDVWM MALDIVES 398 PAKWA WARIDTEL - PAK PAKISTAN

349 MLI01 MALITEL GSM-900 MALI 399 PAKPL ZONG(PAKTEL) PAKISTAN

350 MLI02 ORANGE MALI SA MALI 400 PSEJE JAWWAL(PALCELL) PALESTINE

205



401 PSEWM WATANIYA MOBILE PALESTINE 451 SLEAC AFRICELL-LINTEL SIERRA LEONE

402 PANCW CW PANAMA PANAMA 452 SLECT CELTEL - S.L SIERRA LEONE

403 PANDC DIGICEL PANAMA PANAMA 453 SLECM COMIUM(SL)LMTD. SIERRA LEONE

404 PRYVX HOLA-PARAGUAY PARAGUAY 454 SGPM1 MOBILEONE, SPE SINGAPORE

405 PRYTC TELECEL PARAGUY PARAGUAY 455 SGPST SINGTEL (900) SINGAPORE

406 PERTM CLARO PERU PERU 456 SGPSH STARHUB PTE LT SINGAPORE

407 PHLDG DIGITAL - PHLDG PHILIPPINES 457 SVKGT ORANGE SLOVAK SLOVAKIA

408 PHLGT GLOBE TELECOM PHILIPPINES 458 SVKET T-MOBILE SVK SLOVAKIA

409 PHLSR SMART COMM INC PHILIPPINES 459 SVKO2 TELEFONICA O2 SLOVAKIA

410 POL02 ERA GSM, POLAND POLAND 460 SVNMT MOBITEL,SLOVEN SLOVENIA

411 POLKM POLKOMTEL, POL POLAND 461 SVNSM SI.MOBIL SLOVENIA

412 POL03 PTK-CENTERTEL POLAND 462 SVNVG TUSMOBIL SLOVENIA

413 PRTOP OPTIMUS PORTUGAL 463 SOMNL NATIONLINK TEL. SOMALIA

414 PRTTM TMN, PORTUGAL PORTUGAL 464 ZAFCC CELL C (PTY)LTD SOUTH AFRICA

415 PRTTL VODAFONE - PRT PORTUGAL 465 ZAFMN MTN - S.AFRICA SOUTH AFRICA

416 PRICL CLARO P RICO PUERTO RICO 466 ZAFVC VODACOM - S.A SOUTH AFRICA

417 QATQT QTEL, QATAR QATAR 467 KORKT KT FREETEL SOUTH KOREA

418 QATB1 VODAFONE QATAR QATAR 468 KORKF KT FREETEL CO. SOUTH KOREA

419 REU02 ORANGE REUNION REUNION 469 KORSK SK TELE-S.KOREA SOUTH KOREA

420 REUOT OUTREMER TELCOM REUNION 470 ESPRT ORANGE ES SPAIN

421 FRARE SFR REUNIONAISE REUNION 471 ESPTE TELEFONICA, SPN SPAIN

422 ROMMR ORANGE-ROMANIA ROMANIA 472 ESPAT VODAFONE ESPANA SPAIN

423 ROMCS S.C. COSMOTE ROMANIA 473 LKAAT AIRTEL LANKA SRI LANKA

424 ROMMF VODAFNE ROMANIA ROMANIA 474 LKADG DIALOG TEL PLC SRI LANKA

425 RUSKH BAYKALWESTCOM RUSSIA 475 LKAHT HUTCHISON SRI LANKA

426 RUS17 ERMAK RMS RUSSIA 476 LKA71 MOBITEL-SRILANK SRI LANKA

427 RUSEC LLC EKATERNBURG RUSSIA 477 LKACT TIGO SRI LANKA

428 RUSNW MEGAFON-RUSSIA RUSSIA 478 KNACW ST. KITTS ST. KITTS & NEVIS

429 RUS01 MTS, RUSSIA RUSSIA 479 VCTCW ST.VINCENT ST. VINCENT

430 RUS03 NCC - RUSSIA RUSSIA 480 LCACW S. LUCIA ST.LUCIA

431 RUSNO NOVGOROD TELCOM RUSSIA 481 SDNBT AREEBA SUDAN SUDAN

432 RUS16 NTC - RUSSIA RUSSIA 482 SDNVC VIVACELL SUDAN

433 RUSSC SCS RUSSIA 483 SUDMO ZAIN SD - SUDAN SUDAN

434 RUS14 TELESET LTD. RUSSIA 484 SWZMN MTN - SWAZILAND SWAZILAND

435 RUSUT URALTEL RUSSIA 485 SWEHU HI3G ACCESS-SWE SWEDEN

436 RUSBD VIMPELCOM RUSSIA 486 SWEIQ TELE2, SWEDEN SWEDEN

437 RUS07 ZAO SMARTS RUSSIA 487 SWEEP TELENOR SVERIGE SWEDEN

438 RWAMN MTN RWANDACELL RWANDA 488 SWETR TELIASONERA-SWE SWEDEN

439 RWATG TIGO RWANDA RWANDA 489 CHEOR ORANGE COMM SA SWITZERLAND

440 SMOSM SAN MARINO TELE SAN MARINO 490 CHEDX SUNRISE COMM.AG SWITZERLAND

441 SAUET ETIHAD ETISALAT SAUDI ARABIA 491 CHEC1 SWISSCOM, SWZ SWITZERLAND

442 SAUAJ SAUDI TELECOM SAUDI ARABIA 492 SYRSP AREEBA SYRIA SYRIA

443 SAUZN ZAIN KSA SAUDI ARABIA 493 SYR01 SYRIATEL- SYRIA SYRIA

444 SENSG SENTEL GSM SENEGAL 494 TWNLD CHANGHWA-TAIWAN TAIWAN

445 SENAZ SONATEL SENEGAL 495 TWNFE FAR EAST,TAIWAN TAIWAN

446 YUGTS TELEKOM-SERBIA SERBIA 496 TWNKG KG TELECOM LTD TAIWAN

447 YUGMT TELENOR D.O.O SERBIA 497 TWNPC TAIWAN MBLE CO. TAIWAN

448 SRBNO VIP MOBLE D.O.O SERBIA 498 TWNTG VIBO TELCOM INC TAIWAN

449 SYCAT AIRTEL TELECOM SEYCHELLES 499 TJKBM BABILON-MOBIL TAJIKISTAN

450 SYCCW CABLE WIRELESS SEYCHELLES 500 TJK01 INDIGO NORTH-TJ TAJIKISTAN

206



501 TJKIT INDIGO SOUTH-TJ TAJIKISTAN 551 ZWEET ECONET-ZIMBABWE ZIMBABWE

502 TJK91 TACOM TAJIKISTAN 552 ZWEN1 NET ONE - ZIMB ZIMBABWE

503 TZACT CELTEL - TZA TANZANIA 504 TZAMB MOBITEL-TZA TANZANIA 505 TZAVC VODACOM LTD TANZANIA 506 TZAZN ZANTEL TANZANIA 507 THAAS AIS ,THAILAND THAILAND 508 THAWP TOTAL ACCESS THAILAND 509 THACO TRUE MOVE CO. THAILAND 510 TGOTL TELECEL - TOGO TOGO 511 TTO12 TSTT-TRINIDAD TRINIDAD & TOBAGO 512 TUNOR ORANGE TUNISIA TUNISIA 513 TUNTA TUNISIANA TUNISIA 514 TUNTT TUNISIE TELECOM TUNISIA 515 TURIS AVEA ILETISIM H TURKEY 516 TURTC TURKCELL-TURKEY TURKEY 517 TURTS VODAFONE TELEKO TURKEY 518 TKMBC BCTI TURKMENIST TURKMENISTAN 519 TCACW TURKS & CAICOS TURKS & CAICOS 520 UGACE CELTEL - UGANDA UGANDA 521 UGAMN MTN-UGANDA UGANDA 522 UGAOR ORANGE UGANDA UGANDA 523 UGATL UTL - UGANDA UGANDA 524 UGAWT WARID TELECOM UGANDA 525 GBRHU HUTCHISON 3G-UK UK 526 GBRCN O2 (CELLNET),UK UK 527 GBROR ORANGE, UK UK 528 GBRME T-MOBILE - U.K UK 529 GBRVF VODAFONE, U.K. UK 530 UKRAS ASTELIT UKRAINE 531 UKRGT GOLDEN TELECOM UKRAINE 532 UKRKS KYIVSTAR - GSM UKRAINE 533 UKRUM MOBILECOMM, UKR UKRAINE 534 UKRUT UKRTELECOM UKRAINE 535 UKRRS URS UKRAINE 536 URYAN ANTEL URUGUAY 537 USACG CINGULAR GENES USA 538 USANC NEXTEL COMM USA 539 USAW6 T-MOBILE USA USA 540 UZBDU BELEENE UZ UZBEKISTAN 541 UZB05 UCELL-COSCOM JV UZBEKISTAN 542 UZB07 UZDUNROBITA-UZB UZBEKISTAN 543 VEND2 DIGITEL VENEZUELA 544 VNMVI VIETNAM TELECOM VIETNAM 545 VNMVT VIETTEL TELECM VIETNAM 546 YEMYY HITS-UNITEL YEMEN 547 YEMSA SABAFON YEMEN YEMEN 548 YEMSP SPACETEL (MTN) YEMEN 549 ZMBCE CELTEL - ZAMBIA ZAMBIA 550 ZMB02 MTN (TELECEL) ZAMBIA

THE ECONOMICS OF MOBILE INTERNATIONAL ROAMING · PDF fileTHE ECONOMICS OF MOBILE INTERNATIONAL...

Documents

Transcript of THE ECONOMICS OF MOBILE INTERNATIONAL ROAMING · PDF fileTHE ECONOMICS OF MOBILE INTERNATIONAL...